# Publications

Updated 18 Oct 2016. See also my publications at google, or get a list of all my papers (PDF, updated 23/06/14).

 Authors: ALL AsadaAyBoedeckerLizierMurrayPolaniProkopenkoRiedmillerStolzenburgWangZeman Type: ALL Journal/magazine articlePaper in conference proceedings Publications: selectedALL

### 2016

• Oliver M. Cliff, Joseph T. Lizier, Rosalind X. Wang, Peter Wang, Oliver Obst, and Mikhail Prokopenko. Quantifying Long-Range Interactions and Coherent Structure in Multi-Agent Dynamics. Artificial Life, 2016. To appear (Dec 2016).
[Bibtex] [Abstract] [Details]
We develop and apply several novel methods quantifying dynamic multi-agent team interactions. These interactions are detected information- theoretically and captured in two ways: via (i) directed networks (interaction diagrams) representing significant coupled dynamics between pairs of agents, and (ii) state-space plots (coherence diagrams) showing coherent structures in Shannon information dynamics. This model-free analysis relates, on the one hand, the information transfer to responsiveness of the agents and the team, and, on the other hand, the information storage within the team to the team's rigidity and lack of tactical flexibility. The resultant interaction and coherence diagrams reveal implicit interactions, across teams, that may be spatially long-ranged. The analysis was verified with a statistically significant number of experiments (using simulated football games, produced during RoboCup 2D Simulation League matches), identifying the zones of the most intense competition, the extent and types of interactions, and the correlation between the strength of specific interactions and the results of the matches.

@Article{ CLW+16,
abstract  = { We develop and apply several novel methods quantifying dynamic multi-agent team interactions. These interactions are detected information- theoretically and captured in two ways: via (i) directed networks (interaction diagrams) representing significant coupled dynamics between pairs of agents, and (ii) state-space plots (coherence diagrams) showing coherent structures in Shannon information dynamics. This model-free analysis relates, on the one hand, the information transfer to responsiveness of the agents and the team, and, on the other hand, the information storage within the team to the team's rigidity and lack of tactical flexibility. The resultant interaction and coherence diagrams reveal implicit interactions, across teams, that may be spatially long-ranged. The analysis was verified with a statistically significant number of experiments (using simulated football games, produced during RoboCup 2D Simulation League matches), identifying the zones of the most intense competition, the extent and types of interactions, and the correlation between the strength of specific interactions and the results of the matches.},
author  = {Oliver M. Cliff and Joseph T. Lizier and X. Rosalind Wang and Peter Wang and Oliver Obst and Mikhail Prokopenko},
journal  = {Artificial Life},
keywords  = {multi-agent dynamics, distributed computation, implicit communication, information storage, information transfer},
note = {To appear (Dec 2016).},
title = {Quantifying Long-Range Interactions and Coherent Structure in Multi-Agent Dynamics},
year = {2016}
}

• Ryan Lagerstrom, Yulia Arzhaeva, Piotr Szul, Oliver Obst, Robert Power, Bella Robinson, and Tomasz Bednarz. Image Classification to Support Emergency Situation Awareness. Frontiers in Robotics and AI, 3:54, 2016.
[Bibtex] [Abstract] [Details]
Recent advances in image classification methods, along with the availability of associated tools, has seen their use become widespread in many domains. This paper presents a novel application of current image classification approaches in the area of emergency situation awareness. We discuss image classification based on low level features as well as methods built on top of pre-trained classifiers. The performance of the classifiers are assessed in terms of accuracy along with consideration to computational aspects given the size of the image database. Specifically, we investigate image classification in the context of a bush fire emergency in the Australian state of NSW where images associated with Tweets during the emergency were used to train and test classification approaches. Emergency service operators are interested in having images relevant to such fires reported as extra information to help manage evolving emergencies. We show that these methodologies can classify images into fire and not fire related classes with an accuracy of 86%.

@Article{ LAS+16,
abstract  = {Recent advances in image classification methods, along with the availability of associated tools, has seen their use become widespread in many domains. This paper presents a novel application of current image classification approaches in the area of emergency situation awareness. We discuss image classification based on low level features as well as methods built on top of pre-trained classifiers. The performance of the classifiers are assessed in terms of accuracy along with consideration to computational aspects given the size of the image database. Specifically, we investigate image classification in the context of a bush fire emergency in the Australian state of NSW where images associated with Tweets during the emergency were used to train and test classification approaches. Emergency service operators are interested in having images relevant to such fires reported as extra information to help manage evolving emergencies. We show that these methodologies can classify images into fire and not fire related classes with an accuracy of 86%.},
author  = {Lagerstrom, Ryan and Arzhaeva, Yulia and Szul, Piotr and Obst, Oliver and Power, Robert and Robinson, Bella and Bednarz, Tomasz},
doi = {10.3389/frobt.2016.00054},
journal  = {Frontiers in Robotics and AI},
pages = {54},
title = {Image Classification to Support Emergency Situation Awareness},
volume  = {3},
year = {2016}
}

### 2015

• Mikhail Prokopenko, Lionel Barnett, Michael Harré, Joseph~T. Lizier, Oliver Obst, and X.~Rosalind Wang. Fisher transfer entropy: quantifying the gain in transient sensitivity. Proceedings of the Royal Society of London Series A, 471:20150610, 2015.
[Bibtex] [Abstract] [Details]
We introduce a novel measure, Fisher Transfer Entropy, which quantifies a gain in sensitivity to a control parameter of a state transition, in the context of another observable source. The new measure captures both transient and contextual qualities of transfer entropy and the sensitivity characteristics of Fisher information. Fisher Transfer Entropy is exemplified for a ferromagnetic 2d lattice Ising model with Glauber dynamics, and is shown to diverge at the critical point.

@Article{ PBH+15,
abstract  = {We introduce a novel measure, Fisher Transfer Entropy,
which quantifies a gain in sensitivity to a control parameter of a state transition, in the context of another observable source. The new measure captures both transient and contextual qualities of transfer entropy and the sensitivity characteristics of Fisher information. Fisher Transfer Entropy is exemplified for a ferromagnetic 2d lattice Ising model with Glauber dynamics, and is shown to diverge at the critical point.},
author  = {Prokopenko, Mikhail and Barnett, Lionel and Harr{\'e},
Michael and Lizier, Joseph~T. and Obst, Oliver and Wang,
X.~Rosalind},
doi = {10.1098/rspa.2015.0610},
journal  = {Proceedings of the Royal Society of London Series A},
month = dec,
pages = {20150610},
title = {{Fisher transfer entropy: quantifying the gain in transient sensitivity}},
volume  = {471},
year = {2015}
}

• David M. Budden, Peter Wang, Oliver Obst, and Mikhail Prokopenko. Simulation leagues: Enabling replicable and robust investigation of complex robotic systems. IEEE Robotics and Automation, 22(3):140-146, 2015.
[Bibtex] [Details]
@Article{ BWOP15b,
author  = {David M. Budden and Peter Wang and Oliver Obst and Mikhail Prokopenko},
journal  = {IEEE Robotics and Automation},
keywords  = {simulation; robustness; competitions; ranking},
month = sep,
number  = {3},
pages = {140--146},
title = {Simulation leagues: Enabling replicable and robust investigation of complex robotic systems},
volume  = {22},
year = {2015}
}

• David Budden, Peter Wang, Oliver Obst, and Mikhail Prokopenko. Simulation leagues: Analysis of competition formats. In R. A. C. Bianchi, Levent H. Akin, S. Ramamoorthy, and K. Sugiura, editors, RoboCup 2014: Robot Soccer World Cup XVIII, pages 183-194. Springer, 2015.
[Bibtex] [Abstract] [Details]
A preprint can be found on arxiv.org/abs/1403.4023.

The selection of an appropriate competition format is critical for both the success and credibility of any competition, both real and simulated. In this paper, the automated parallelism offered by the RoboCupSoccer 2D simulation league is leveraged to conduct a 28,000 game round-robin between the top 8 teams from RoboCup 2012 and 2013. A proposed new competition format is found to reduce variation from the resultant statistically significant team performance rankings by 75 % and 67 %, when compared to the actual competition results from RoboCup 2012 and 2013 respectively. These results are statistically validated by generating 10,000 random tournaments for each of the three considered formats and comparing the respective distributions of ranking discrepancy.

@InCollection{ BWOP15a,
abstract  = {The selection of an appropriate competition format is critical for both the success and credibility of any competition, both real and simulated. In this paper, the automated parallelism offered by the RoboCupSoccer 2D simulation league is leveraged to conduct a 28,000 game round-robin between the top 8 teams from RoboCup 2012 and 2013. A proposed new competition format is found to reduce variation from the resultant statistically significant team performance rankings by 75 % and 67 %, when compared to the actual competition results from RoboCup 2012 and 2013 respectively. These results are statistically validated by generating 10,000 random tournaments for each of the three considered formats and comparing the respective distributions of ranking discrepancy.},
author  = {David Budden and Peter Wang and Oliver Obst and Mikhail Prokopenko},
booktitle  = {{R}obo{C}up 2014: Robot Soccer World Cup {XVIII}},
editor  = {R. A. C. Bianchi and H. Levent Akin and S. Ramamoorthy and K Sugiura},
keywords  = {robotics; simulation; soccer matches; multiagent systems; artificial intelligence},
pages = {183--194},
publisher  = {Springer},
title = {Simulation leagues: Analysis of competition formats},
wwwnote  = {A preprint can be found on <a href="http://arxiv.org/abs/1403.4023">arxiv.org/abs/1403.4023</a>.}
,
year = {2015}
}

### 2014

• Astrid Zeman, Oliver Obst, and Kevin R. Brooks. Complex cells decrease errors for the Müller-Lyer illusion in a model of the visual ventral stream. Frontiers in Computational Neuroscience, 8(112), 2014.
[Bibtex] [Abstract] [Details]
To improve robustness in object recognition, many artificial visual systems imitate the way in which the human visual cortex encodes object information as a hierarchical set of features. These systems are usually evaluated in terms of their ability to accurately categorize well-defined, unambiguous objects and scenes. In the real world, however, not all objects and scenes are presented clearly, with well-defined labels and interpretations. Visual illusions demonstrate a disparity between perception and objective reality, allowing psychophysicists to methodically manipulate stimuli and study our interpretation of the environment. One prominent effect, the Müller-Lyer illusion, is demonstrated when the perceived length of a line is contracted (or expanded) by the addition of arrowheads (or arrow-tails) to its ends. HMAX, a benchmark object recognition system, consistently produces a bias when classifying Müller-Lyer images. HMAX is a hierarchical, artificial neural network that imitates the simple'' and complex'' cell layers found in the visual ventral stream. In this study, we perform two experiments to explore the Müller-Lyer illusion in HMAX, asking: (1) How do simple vs. complex cell operations within HMAX affect illusory bias and precision? (2) How does varying the position of the figures in the input image affect classification using HMAX? In our first experiment, we assessed classification after traversing each layer of HMAX and found that in general, kernel operations performed by simple cells increase bias and uncertainty while max-pooling operations executed by complex cells decrease bias and uncertainty. In our second experiment, we increased variation in the positions of figures in the input images that reduced bias and uncertainty in HMAX. Our findings suggest that the Müller-Lyer illusion is exacerbated by the vulnerability of simple cell operations to positional fluctuations, but ameliorated by the robustness of complex cell responses to such variance.

@Article{ ZOB14,
abstract  = {To improve robustness in object recognition, many artificial visual systems imitate the way in which the human visual cortex encodes object information as a hierarchical set of features. These systems are usually evaluated in terms of their ability to accurately categorize well-defined, unambiguous objects and scenes. In the real world, however, not all objects and scenes are presented clearly, with well-defined labels and interpretations. Visual illusions demonstrate a disparity between perception and objective reality, allowing psychophysicists to methodically manipulate stimuli and study our interpretation of the environment. One prominent effect, the M{\"u}ller-Lyer illusion, is demonstrated when the perceived length of a line is contracted (or expanded) by the addition of arrowheads (or arrow-tails) to its ends. HMAX, a benchmark object recognition system, consistently produces a bias when classifying M{\"u}ller-Lyer images. HMAX is a hierarchical, artificial neural network that imitates the simple'' and complex'' cell layers found in the visual ventral stream. In this study, we perform two experiments to explore the M{\"u}ller-Lyer illusion in HMAX, asking: (1) How do simple vs. complex cell operations within HMAX affect illusory bias and precision? (2) How does varying the position of the figures in the input image affect classification using HMAX? In our first experiment,
we assessed classification after traversing each layer of HMAX and found that in general, kernel operations performed by simple cells increase bias and uncertainty while max-pooling operations executed by complex cells decrease bias and uncertainty. In our second experiment, we increased variation in the positions of figures in the input images that reduced bias and uncertainty in HMAX. Our findings suggest that the M{\"u}ller-Lyer illusion is exacerbated by the vulnerability of simple cell operations to positional fluctuations, but ameliorated by the robustness of complex cell responses to such variance.},
author  = {Zeman, Astrid and Obst, Oliver and Brooks, Kevin R.},
doi = {10.3389/fncom.2014.00112},
issn = {1662-5188},
journal  = {Frontiers in Computational Neuroscience},
number  = {112},
title = {Complex cells decrease errors for the M{\"u}ller-Lyer illusion in a model of the visual ventral stream},
url = {http://www.frontiersin.org/computational_neuroscience/10.3389/fncom.2014.00112/abstract}
,
volume  = {8},
year = {2014}
}

• Oliver Obst and Joschka Boedecker. Guided Self-Organization: Inception, volume 9 of Emergence, Complexity and Computation, chapter Guided Self-Organization of Input-Driven Recurrent Neural Networks, pages 319-340. Springer, 2014.
[Bibtex] [Details]
At Springer; a preprint at http://arxiv.org/abs/1309.1524.
@InBook{ OB14,
author  = {Oliver Obst and Joschka Boedecker},
chapter  = {Guided Self-Organization of Input-Driven Recurrent Neural Networks},
doi = {10.1007/978-3-642-53734-9_11},
editor  = {Mikhail Prokopenko},
pages = {319--340},
publisher  = {Springer},
series  = {Emergence, Complexity and Computation},
title = {Guided Self-Organization: Inception},
url = {http://arxiv.org/abs/1309.1524},
volume  = {9},
wwwnote  = {At <a href="http://link.springer.com/chapter/10.1007%2F978-3-642-53734-9_11">Springer</a>; a preprint at <a href="http://arxiv.org/abs/1309.1524">http://arxiv.org/abs/1309.1524</a>.}
,
year = {2014}
}

• Oliver Obst. Distributed machine learning and sparse representations. Neurocomputing, 124:1, 2014.
[Bibtex] [Abstract] [Details]
The need to process large amounts of data has led to a wave of developments concerned with the theory and practice of knowledge discovery and machine learning. In particular, social networks and web applications contributed to the explosion of data and an increased need for faster and scalable methods. Similarly, though maybe less obviously, large amounts of data started to play an important role in areas outside the social web. An increasing proliferation of sensors deployed in the environment, embedded in devices, vehicles and buildings result in continuous streams of data. The information we are able to get from these streams is as rich as the information we get from social webs, but it may be of a different nature -- e.g., environmental data, high-dimensional traffic data or internal sensors of a robot or a vehicle -- and consequently call for different approaches.

@Article{ Obst14a,
abstract  = {The need to process large amounts of data has led to a wave of developments concerned with the theory and practice of knowledge discovery and machine learning. In particular,
social networks and web applications contributed to the explosion of data and an increased need for faster and scalable methods. Similarly, though maybe less obviously,
large amounts of data started to play an important role in areas outside the social web. An increasing proliferation of sensors deployed in the environment, embedded in devices, vehicles and buildings result in continuous streams of data. The information we are able to get from these streams is as rich as the information we get from social webs, but it may be of a different nature -- e.g.,
environmental data, high-dimensional traffic data or internal sensors of a robot or a vehicle -- and consequently call for different approaches. },
author  = {Oliver Obst},
doi = {10.1016/j.neucom.2013.03.009},
journal  = {Neurocomputing},
pages = {1},
title = {Distributed machine learning and sparse representations},
volume  = {124},
year = {2014}
}

• Oliver Obst. Distributed Fault Detection in Sensor Networks Using a Recurrent Neural Network. Neural Processing Letters, 40(3):261-273, 2014.
[Bibtex] [Abstract] [Details]
Published online ahead of print: Online First. A preprint can be found at http://arxiv.org/abs/0906.4154.

In long-term deployments of sensor networks, monitoring the quality of gathered data is a critical issue. Over the time of deployment, sensors are exposed to harsh conditions, causing some of them to fail or to deliver less accurate data. If such a degradation remains undetected, the usefulness of a sensor network can be greatly reduced. We present an approach that learns spatio-temporal correlations between different sensors, and makes use of the learned model to detect anomalous sensors by using distributed computation and only local communication between nodes. We introduce SODESN, a distributed recurrent neural network architec- ture, and a learning method to train SODESN for fault detection in a distributed scenario. Our approach is evaluated using data from a real-world sensor-network deployment, and shows good results even with imperfect link qualities and a num- ber of simultaneous faults.

@Article{ Obst14b,
abstract  = {In long-term deployments of sensor networks, monitoring the quality of gathered data is a critical issue. Over the time of deployment, sensors are exposed to harsh conditions, causing some of them to fail or to deliver less accurate data. If such a degradation remains undetected,
the usefulness of a sensor network can be greatly reduced. We present an approach that learns spatio-temporal correlations between different sensors, and makes use of the learned model to detect anomalous sensors by using distributed computation and only local communication between nodes. We introduce SODESN, a distributed recurrent neural network architec- ture, and a learning method to train SODESN for fault detection in a distributed scenario. Our approach is evaluated using data from a real-world sensor-network deployment, and shows good results even with imperfect link qualities and a num- ber of simultaneous faults.},
author  = {Oliver Obst},
journal  = {Neural Processing Letters},
keywords  = {Neural networks, reservoir computing, sensor networks,
fault detection},
number  = {3},
pages = {261--273},
title = {Distributed Fault Detection in Sensor Networks Using a Recurrent Neural Network},
volume  = {40},
wwwnote  = {Published online ahead of print: <a href="http://link.springer.com/article/10.1007%2Fs11063-013-9327-4">Online First</a>. A preprint can be found at <a href="http://arxiv.org/abs/0906.4154">http://arxiv.org/abs/0906.4154</a>.}
,
year = {2014}
}

### 2013

• Oliver Obst, Adrian Trinchi, Simon G. Hardin, Matthew Chadwick, Ivan Cole, Tim H. Muster, Nigel Hoschke, Diet Ostry, Don Price, Khoa N. Pham, and Tim Wark. Nano-scale reservoir computing. Nano Communication Networks, 4(4):189-196, 2013.
[Bibtex] [Abstract] [Details]
See also http://arxiv.org/abs/1309.1521. An earlier version appeared in the Proceedings of the 3rd IEEE International Workshop on Molecular and Nanoscale Communications (IEEE MoNaCom 2013) and can be downloaded at http://www.oliverobst.eu/publications/2013/OTH+13.pdf.

This work describes preliminary steps towards nano-scale reservoir computing using quantum dots. Our research has focused on the development of an accumulator-based sensing system that reacts to changes in the environment, as well as the development of a software simulation. The investigated systems generate nonlinear responses to inputs that make them suitable for a physical implementation of a neural network. This development will enable miniaturisation of the neurons to the molecular level, leading to a range of applications including monitoring of changes in materials or structures. The system is based around the optical properties of quantum dots. The paper will report on experimental work on systems using Cadmium Selenide (CdSe) quantum dots and on the various methods to render the systems sensitive to pH, redox potential or specific ion concentration. Once the quantum dot-based systems are rendered sensitive to these triggers they can provide a distributed array that can monitor and transmit information on changes within the material.

@Article{ OTH+13b,
abstract  = {This work describes preliminary steps towards nano-scale reservoir computing using quantum dots. Our research has focused on the development of an accumulator-based sensing system that reacts to changes in the environment, as well as the development of a software simulation. The investigated systems generate nonlinear responses to inputs that make them suitable for a physical implementation of a neural network. This development will enable miniaturisation of the neurons to the molecular level,
leading to a range of applications including monitoring of changes in materials or structures. The system is based around the optical properties of quantum dots. The paper will report on experimental work on systems using Cadmium Selenide (CdSe) quantum dots and on the various methods to render the systems sensitive to pH, redox potential or specific ion concentration. Once the quantum dot-based systems are rendered sensitive to these triggers they can provide a distributed array that can monitor and transmit information on changes within the material.},
author  = {Oliver Obst and Adrian Trinchi and Simon G. Hardin and Matthew Chadwick and Ivan Cole and Tim H. Muster and Nigel Hoschke and Diet Ostry and Don Price and Khoa N. Pham and Tim Wark},
journal  = {Nano Communication Networks},
keywords  = {Quantum dots, recurrent neural networks, Echo State Networks},
number  = {4},
pages = {189--196},
title = {Nano-scale reservoir computing},
url = {http://www.sciencedirect.com/science/article/pii/S1878778913000495}
,
volume  = {4},
wwwnote  = {See also <a href="http://arxiv.org/abs/1309.1521">http://arxiv.org/abs/1309.1521</a>. An <a href="/publications/2013/OTH+13.pdf">earlier version</a> appeared in the Proceedings of the 3rd IEEE International Workshop on Molecular and Nanoscale Communications (IEEE MoNaCom 2013) and can be downloaded at <a href="/publications/2013/OTH+13.pdf">http://www.oliverobst.eu/publications/2013/OTH+13.pdf</a>.}
,
year = {2013}
}

• Oliver M. Cliff, Joseph T. Lizier, Rosalind X. Wang, Peter Wang, Oliver Obst, and Mikhail Prokopenko. Towards Quantifying Interaction Networks in a Football Match. In Proceedings of the RoboCup 2013 Symposium, 2013.
[Bibtex] [Abstract] [Details]
Best paper award for the best theoretical contribution.

We present several novel methods quantifying dynamic interactions in simulated football games. These interactions are captured in directed networks that represent significant coupled dynamics, detected information-theoretically. The model-free approach measures information dynamics of both pair-wise players' interactions as well as local tactical contests produced during RoboCup 2D Simulation League games. This analysis involves computation of information transfer and storage, relating the information transfer to responsiveness of the players and the team, and the information storage within the team to the team's rigidity and lack of tactical flexibility. The resultant directed networks (interaction diagrams) and the measures of responsiveness and rigidity reveal implicit interactions, across teams, that may be delayed and/or long-ranged. The analysis was verified with a number of experiments, identifying the zones of the most intense competition and the extent of interactions.

@InProceedings{ CLW+13,
abstract  = {We present several novel methods quantifying dynamic interactions in simulated football games. These interactions are captured in directed networks that represent significant coupled dynamics, detected information-theoretically. The model-free approach measures information dynamics of both pair-wise players' interactions as well as local tactical contests produced during RoboCup 2D Simulation League games. This analysis involves computation of information transfer and storage,
relating the information transfer to responsiveness of the players and the team, and the information storage within the team to the team's rigidity and lack of tactical flexibility. The resultant directed networks (interaction diagrams) and the measures of responsiveness and rigidity reveal implicit interactions, across teams, that may be delayed and/or long-ranged. The analysis was verified with a number of experiments, identifying the zones of the most intense competition and the extent of interactions.},
author  = {Oliver M. Cliff and Joseph T. Lizier and X. Rosalind Wang and Peter Wang and Oliver Obst and Mikhail Prokopenko},
booktitle  = {Proceedings of the {RoboCup 2013} Symposium},
title = {Towards Quantifying Interaction Networks in a Football Match},
wwwnote  = {<strong>Best paper award for the best theoretical contribution.</strong>},
year = {2013}
}

• Oliver Obst, Joschka Boedecker, Benedikt Schmidt, and Minoru Asada. On active information storage in input-driven systems. arXiv preprint 1303.5526v1, arXiv.org, 2013. Also published as a poster at the 6th Australian Workshop on Computational Neuroscience, 2013
[Bibtex] [Abstract] [Details]
Information theory and the framework of information dynamics have been used to provide tools to characterise complex systems. In particular, we are interested in quantifying information storage, information modification and information transfer as characteristic elements of computation. Although these quantities are defined for autonomous dynamical systems, information dynamics can also help to get a wholistic'' understanding of input-driven systems such as neural networks. In this case, we do not distinguish between the system itself, and the effects the input has to the system. This may be desired in some cases, but it will change the questions we are able to answer, and is consequently an important consideration, for example, for biological systems which perform non-trivial computations and also retain a short-term memory of past inputs.Many other real world systems like cortical networks are also heavily input-driven, and application of tools designed for autonomous dynamic systems may not necessarily lead to intuitively interpretable results. The aim of our work is to extend the measurements used in the information dynamics framework for input-driven systems. Using the proposed input-corrected information storage we hope to better quantify system behaviour, which will be important for heavily input-driven systems like artificial neural networks to abstract from specific benchmarks, or for brain networks, where intervention is difficult, individual components cannot be tested in isolation or with arbitrary input data.

@TechReport{ OBSA13b,
abstract  = {Information theory and the framework of information dynamics have been used to provide tools to characterise complex systems. In particular, we are interested in quantifying information storage, information modification and information transfer as characteristic elements of computation. Although these quantities are defined for autonomous dynamical systems, information dynamics can also help to get a wholistic'' understanding of input-driven systems such as neural networks. In this case, we do not distinguish between the system itself, and the effects the input has to the system. This may be desired in some cases,
but it will change the questions we are able to answer, and is consequently an important consideration, for example,
for biological systems which perform non-trivial computations and also retain a short-term memory of past inputs.Many other real world systems like cortical networks are also heavily input-driven, and application of tools designed for autonomous dynamic systems may not necessarily lead to intuitively interpretable results. The aim of our work is to extend the measurements used in the information dynamics framework for input-driven systems. Using the proposed input-corrected information storage we hope to better quantify system behaviour, which will be important for heavily input-driven systems like artificial neural networks to abstract from specific benchmarks, or for brain networks, where intervention is difficult, individual components cannot be tested in isolation or with arbitrary input data.},
author  = {Oliver Obst and Joschka Boedecker and Benedikt Schmidt and Minoru Asada},
institution  = {arXiv.org},
note = {Also published as a poster at the 6th Australian Workshop on Computational Neuroscience, 2013},
number  = {1303.5526v1},
title = {On active information storage in input-driven systems},
type = {arXiv preprint},
url = {http://arxiv.org/abs/1303.5526},
year = {2013}
}

• Astrid Zeman, Oliver Obst, Kevin R. Brooks, and Anina N. Rich. The Müller-Lyer Illusion in a Computational Model of Biological Object Recognition. PLoS ONE, 8(2):e56126, 2013.
[Bibtex] [Abstract] [Details]
Studying illusions provides insight into the way the brain processes information. The Müller-Lyer Illusion (MLI) is a classical geometrical illusion of size, in which perceived line length is decreased by arrowheads and increased by arrowtails. Many theories have been put forward to explain the MLI, such as misapplied size constancy scaling, the statistics of image-source relationships and the filtering properties of signal processing in primary visual areas. Artificial models of the ventral visual processing stream allow us to isolate factors hypothesised to cause the illusion and test how these affect classification performance. We trained a feed-forward feature hierarchical model, HMAX, to perform a dual category line length judgment task (short versus long) with over 90% accuracy. We then tested the system in its ability to judge relative line lengths for images in a control set versus images that induce the MLI in humans. Results from the computational model show an overall illusory effect similar to that experienced by human subjects. No natural images were used for training, implying that misapplied size constancy and image-source statistics are not necessary factors for generating the illusion. A post-hoc analysis of response weights within a representative trained network ruled out the possibility that the illusion is caused by a reliance on information at low spatial frequencies. Our results suggest that the MLI can be produced using only feed-forward, neurophysiological connections.

@Article{ ZOBR13,
abstract  = {Studying illusions provides insight into the way the brain processes information. The M{\"u}ller-Lyer Illusion (MLI) is a classical geometrical illusion of size, in which perceived line length is decreased by arrowheads and increased by arrowtails. Many theories have been put forward to explain the MLI, such as misapplied size constancy scaling, the statistics of image-source relationships and the filtering properties of signal processing in primary visual areas. Artificial models of the ventral visual processing stream allow us to isolate factors hypothesised to cause the illusion and test how these affect classification performance. We trained a feed-forward feature hierarchical model, HMAX, to perform a dual category line length judgment task (short versus long) with over 90% accuracy. We then tested the system in its ability to judge relative line lengths for images in a control set versus images that induce the MLI in humans. Results from the computational model show an overall illusory effect similar to that experienced by human subjects. No natural images were used for training,
implying that misapplied size constancy and image-source statistics are not necessary factors for generating the illusion. A post-hoc analysis of response weights within a representative trained network ruled out the possibility that the illusion is caused by a reliance on information at low spatial frequencies. Our results suggest that the MLI can be produced using only feed-forward, neurophysiological connections.},
author  = {Astrid Zeman and Oliver Obst and Kevin R. Brooks and Anina N. Rich},
doi = {10.1371/journal.pone.0056126},
journal  = {PLoS ONE},
month = feb,
number  = {2},
pages = {e56126},
title = {The M{\"u}ller-Lyer Illusion in a Computational Model of Biological Object Recognition},
volume  = {8},
year = {2013}
}

### 2012

• Christoph Hartmann, Joschka Boedecker, Oliver Obst, Shuhei Ikemoto, and Minoru Asada. Real-Time Inverse Dynamics Learning for Musculoskeletal Robots based on Echo State Gaussian Process Regression. In Robotics: Science and Systems (RSS 2012), 2012.
[Bibtex] [Abstract] [Details]
A challenging topic in articulated robots is the control of redundantly many degrees of freedom with artificial muscles. Actuation with these devices is difficult to solve because of nonlinearities, delays and unknown parameters such as friction that are involved. Machine learning methods can be used to learn the control of these systems, but they are faced with the additional problem that the size of the search space prohibits full exploration in reasonable time. In this paper, we propose a novel method that is able to learn control of redundant robot arms with artificial muscles online from scratch using only the position of the end effector, without using any joint positions, accelerations or an analytical model of the system or the environment. To be able to learn in real time, we use the so called online goal babbling'' method to effectively reduce the search space, a recurrent neural network to represent the state of the robot arm, and novel online Gaussian processes for regression. With our approach, we achieve good performance on trajectory tracking tasks for the end effector of two very challenging systems: a simulated 6 DOF redundant arm with artificial muscles, and a 7 DOF robot arm with McKibben pneumatic artificial muscles. We also show that the combination of techniques we propose results in significantly improved performance over using the individual techniques alone.

@InProceedings{ HBO+12,
abstract  = {A challenging topic in articulated robots is the control of redundantly many degrees of freedom with artificial muscles. Actuation with these devices is difficult to solve because of nonlinearities, delays and unknown parameters such as friction that are involved. Machine learning methods can be used to learn the control of these systems,
but they are faced with the additional problem that the size of the search space prohibits full exploration in reasonable time. In this paper, we propose a novel method that is able to learn control of redundant robot arms with artificial muscles online from scratch using only the position of the end effector, without using any joint positions, accelerations or an analytical model of the system or the environment. To be able to learn in real time, we use the so called online goal babbling'' method to effectively reduce the search space, a recurrent neural network to represent the state of the robot arm, and novel online Gaussian processes for regression. With our approach, we achieve good performance on trajectory tracking tasks for the end effector of two very challenging systems: a simulated 6 DOF redundant arm with artificial muscles, and a 7 DOF robot arm with McKibben pneumatic artificial muscles. We also show that the combination of techniques we propose results in significantly improved performance over using the individual techniques alone. },
author  = {Christoph Hartmann and Joschka Boedecker and Oliver Obst and Shuhei Ikemoto and Minoru Asada},
booktitle  = {Robotics: Science and Systems (RSS 2012)},
title = {Real-Time Inverse Dynamics Learning for Musculoskeletal Robots based on Echo State Gaussian Process Regression},
year = {2012}
}

• Oliver Obst and Martin Riedmiller. Taming the Reservoir: Feedforward Training for Recurrent Neural Networks. In The International Joint Conference on Neural Networks (IJCNN 2012), pages 1-7. IEEE, 2012.
[Bibtex] [Abstract] [Details]
Nominated for a best paper award.

Recurrent neural networks are successfully used for tasks like time series processing and system identification. Many of the approaches to train these networks, however, are often regarded as too slow, too complicated, or both. Reservoir computing methods like echo state networks or liquid state machines are an alternative to the more traditional approaches. Echo state networks have the appeal that they are simple to train, and that they have shown to be able to produce excellent results for a number of benchmarks and other tasks. One disadvantage of echo state networks however is the high variability in their performance due to a randomly connected hidden layer. Ideally, an efficient and more deterministic way to create connections in the hidden layer could be found, with a performance better than randomly connected hidden layers but without excessively iterating over the same training data many times. We present an approach that makes use of efficient feedforward training methods, and performs better than echo state networks for some time series prediction tasks. Moreover, our approach reduces some of the variability since all recurrent connections in the network are trained.

@InProceedings{ OR12,
abstract  = {Recurrent neural networks are successfully used for tasks like time series processing and system identification. Many of the approaches to train these networks, however, are often regarded as too slow, too complicated, or both. Reservoir computing methods like echo state networks or liquid state machines are an alternative to the more traditional approaches. Echo state networks have the appeal that they are simple to train, and that they have shown to be able to produce excellent results for a number of benchmarks and other tasks. One disadvantage of echo state networks however is the high variability in their performance due to a randomly connected hidden layer. Ideally, an efficient and more deterministic way to create connections in the hidden layer could be found, with a performance better than randomly connected hidden layers but without excessively iterating over the same training data many times. We present an approach that makes use of efficient feedforward training methods, and performs better than echo state networks for some time series prediction tasks. Moreover, our approach reduces some of the variability since all recurrent connections in the network are trained.},
author  = {Oliver Obst and Martin Riedmiller},
booktitle  = {The International Joint Conference on Neural Networks (IJCNN 2012)},
pages = {1--7},
publisher  = {IEEE},
title = {Taming the Reservoir: Feedforward Training for Recurrent Neural Networks},
wwwnote  = {<strong>Nominated for a best paper award.</strong>},
year = {2012}
}

• Joschka Boedecker, Oliver Obst, Joseph T. Lizier, Michael N. Mayer, and Minoru Asada. Information Processing in Echo State Networks at the Edge of Chaos. Theory In Biosciences, 131(3):205-213, sep 2012.
[Bibtex] [Abstract] [Details]
We investigate information processing in randomly connected recurrent neural networks. It has been shown previously that the computational capabilities of these networks are maximized when the recurrent layer is close to the border between a stable and an unstable dynamics regime, the so called \emph{edge of chaos}. The reasons, however, for this maximized performance are not completely understood. We adopt an information-theoretical framework and are for the first time able to quantify the computational capabilities between elements of these networks directly as they undergo the phase transition to chaos. Specifically, we present evidence that both information transfer and storage in the recurrent layer peak close to this phase transition, providing an explanation for why guiding the recurrent layer towards the edge of chaos is computationally useful. As a consequence, our work suggests self-organized ways of improving performance in recurrent neural networks, driven by input data. Moreover, the networks we study share important features with biological systems such as feedback connections and online computation on input streams. A key example is the cerebral cortex, which was shown to also operate close to the edge of chaos. Consequently, the behavior of model systems as studied here is likely to shed light on reasons why biological systems are tuned into this specific regime.

@Article{ BOL+12,
abstract  = {We investigate information processing in randomly connected recurrent neural networks. It has been shown previously that the computational capabilities of these networks are maximized when the recurrent layer is close to the border between a stable and an unstable dynamics regime, the so called \emph{edge of chaos}. The reasons, however, for this maximized performance are not completely understood. We adopt an information-theoretical framework and are for the first time able to quantify the computational capabilities between elements of these networks directly as they undergo the phase transition to chaos. Specifically, we present evidence that both information transfer and storage in the recurrent layer peak close to this phase transition,
providing an explanation for why guiding the recurrent layer towards the edge of chaos is computationally useful. As a consequence, our work suggests self-organized ways of improving performance in recurrent neural networks, driven by input data. Moreover, the networks we study share important features with biological systems such as feedback connections and online computation on input streams. A key example is the cerebral cortex, which was shown to also operate close to the edge of chaos. Consequently, the behavior of model systems as studied here is likely to shed light on reasons why biological systems are tuned into this specific regime. },
author  = {Joschka Boedecker and Oliver Obst and Joseph T. Lizier and Mayer, N. Michael and Minoru Asada},
journal  = {Theory In Biosciences},
keywords  = {Recurrent Neural Networks, Reservoir Computing,
Information Transfer, Active Information Storage, Phase Transition},
month = sep,
number  = {3},
pages = {205--213},
title = {Information Processing in Echo State Networks at the Edge of Chaos},
volume  = {131},
year = {2012}
}

### 2011

• Mikhail Prokopenko, Joseph T. Lizier, Oliver Obst, and Rosalind X. Wang. Relating Fisher information to order parameters. Physical Review E, 84(4):41116, 2011.
[Bibtex] [Abstract] [Details]
We study phase transitions and relevant order parameters via statistical estimation theory using the Fisher information matrix. The assumptions that we make limit our analysis to order parameters representable as a negative derivative of thermodynamic potential over some thermodynamic vari- able. Nevertheless, the resulting representation is sufficiently general and explicitly relates elements of the Fisher information matrix to the rate of change in the corresponding order parameters. The obtained relationships allow us to identify, in particular, second-order phase transitions via diver- gences of individual elements of the Fisher information matrix. A computational study of random Boolean networks (RBNs) supports the derived relationships, illustrating that Fisher information of the magnetization bias (that is, activity level) is peaked in finite-size networks at the critical points, and the maxima increase with the network size. The framework presented here reveals the basic thermodynamic reasons behind similar empirical observations reported previously. The study highlights the generality of Fisher information as a measure that can be applied to a broad range of systems, particularly those where the determination of order parameters is cumbersome.

@Article{ PLOW11,
abstract  = {We study phase transitions and relevant order parameters via statistical estimation theory using the Fisher information matrix. The assumptions that we make limit our analysis to order parameters representable as a negative derivative of thermodynamic potential over some thermodynamic vari- able. Nevertheless, the resulting representation is sufficiently general and explicitly relates elements of the Fisher information matrix to the rate of change in the corresponding order parameters. The obtained relationships allow us to identify, in particular,
second-order phase transitions via diver- gences of individual elements of the Fisher information matrix. A computational study of random Boolean networks (RBNs) supports the derived relationships, illustrating that Fisher information of the magnetization bias (that is,
activity level) is peaked in finite-size networks at the critical points, and the maxima increase with the network size. The framework presented here reveals the basic thermodynamic reasons behind similar empirical observations reported previously. The study highlights the generality of Fisher information as a measure that can be applied to a broad range of systems, particularly those where the determination of order parameters is cumbersome.},
author  = {Mikhail Prokopenko and Joseph T. Lizier and Oliver Obst and X. Rosalind Wang},
doi = {10.1103/PhysRevE.84.041116},
journal  = {Physical Review E},
keywords  = {phase transitions, Fisher information, order parameter,
thermodynamic potential, free entropy, random Boolean network, critical points},
number  = {4},
pages = {041116},
title = {Relating Fisher information to order parameters},
volume  = {84},
year = {2011}
}

• Astrid Zeman, Oliver Obst, and Anina N. Rich. Exploring the Müller-Lyer illusion using an artificial feed-forward object recognition model. Perception, 40 ECVP Abstract Supplement:201, 2011.
[Bibtex] [Abstract] [Details]
Best student poster, European Conference on Visual Perception (Toulouse, France, 2011). Presenter: A. Zeman

The Müller-Lyer (ML) illusion is where perceived line length is decreased by inward arrowheads, but is increased by outward arrowheads. Many theories have been put forward to explain the ML illusion, such as filtering properties of signal processing in primary visual areas (Bulatov et al 1997). Artificial models of the ventral visual processing stream provide us with the potential to isolate and test how exposure to different image sets affects classification performance. We trained a feed-forward hierarchical model (Mutch & Lowe, 2008) to perform a dual category line length judgment task (short versus long) with over 90% accuracy. We trained the model using a control set of images that would capture features present in illusion stimuli. We then tested the system in its ability to judge relative line lengths for images in a control set versus images that induce the ML illusion in humans. In this way, we were able to isolate and observe the effect of exposure to different stimuli on illusion judgment in a simple-complex feed-forward network.

@Article{ ZOR11,
abstract  = {The M{\"u}ller-Lyer (ML) illusion is where perceived line length is decreased by inward arrowheads, but is increased by outward arrowheads. Many theories have been put forward to explain the ML illusion, such as filtering properties of signal processing in primary visual areas (Bulatov et al 1997). Artificial models of the ventral visual processing stream provide us with the potential to isolate and test how exposure to different image sets affects classification performance. We trained a feed-forward hierarchical model (Mutch & Lowe,
2008) to perform a dual category line length judgment task (short versus long) with over 90% accuracy. We trained the model using a control set of images that would capture features present in illusion stimuli. We then tested the system in its ability to judge relative line lengths for images in a control set versus images that induce the ML illusion in humans. In this way, we were able to isolate and observe the effect of exposure to different stimuli on illusion judgment in a simple-complex feed-forward network.},
author  = {Astrid Zeman and Oliver Obst and Anina N. Rich},
doi = {10.1068/v110367},
journal  = {Perception},
month = aug,
pages = {201},
title = {Exploring the {M\"uller-Lyer} illusion using an artificial feed-forward object recognition model},
volume  = {40 ECVP Abstract Supplement},
wwwnote  = {<strong>Best student poster, European Conference on Visual Perception (Toulouse, France, 2011). Presenter: A. Zeman</strong>},
year = {2011}
}

• Raja Jurdak, Rosalind X. Wang, Oliver Obst, and Philip Valencia. Wireless Sensor Network Anomalies: Diagnosis and Detection Strategies, volume 10 of Intelligent Systems Reference Library, chapter 12, pages 309-325. Springer, Berlin, Heidelberg, 2011.
[Bibtex] [Abstract] [Details]
Wireless Sensor Networks (WSNs) can experience problems (anomalies) during deployment, due to dynamic environmental factors or node hardware and software failures. These anomalies demand reliable detection strategies for supporting long term and/or large scale WSN deployments. Several strategies have been proposed for detecting specific subsets of WSN anomalies, yet there is still a need for more comprehensive anomaly detection strategies that jointly address network, node, and data level anomalies. This chapter examines WSN anomalies from an intelligent-based system perspective, covering anomalies that arise at the network, node and data levels. It generalizes a simple process for diagnosing anomalies in WSNs for detection, localization, and root cause determination. A survey of existing anomaly detection strategies also reveals their major design choices, including architecture and user support, and yields guidelines for tailoring new anomaly detection strategies to specific WSN application requirements.

@InBook{ JWOV11,
abstract  = {Wireless Sensor Networks (WSNs) can experience problems (anomalies) during deployment, due to dynamic environmental factors or node hardware and software failures. These anomalies demand reliable detection strategies for supporting long term and/or large scale WSN deployments. Several strategies have been proposed for detecting specific subsets of WSN anomalies, yet there is still a need for more comprehensive anomaly detection strategies that jointly address network, node, and data level anomalies. This chapter examines WSN anomalies from an intelligent-based system perspective, covering anomalies that arise at the network, node and data levels. It generalizes a simple process for diagnosing anomalies in WSNs for detection, localization, and root cause determination. A survey of existing anomaly detection strategies also reveals their major design choices,
including architecture and user support, and yields guidelines for tailoring new anomaly detection strategies to specific WSN application requirements.},
affiliation  = {CSIRO ICT Centre, Australia},
author  = {Jurdak, Raja and Wang, X. Rosalind and Obst, Oliver and Valencia, Philip},
booktitle  = {Intelligence-Based Systems Engineering},
chapter  = {12},
doi = {10.1007/978-3-642-17931-0_12},
editor  = {Kacprzyk, Janusz and Jain, Lakhmi C. and Tolk, Andreas and Jain, Lakhmi C.},
isbn = {978-3-642-17931-0},
keyword  = {wireless sensor networks, anomaly detection, diagnosis},
pages = {309--325},
publisher  = {Springer},
series  = {Intelligent Systems Reference Library},
title = {Wireless Sensor Network Anomalies: Diagnosis and Detection Strategies},
volume  = {10},
year = {2011}
}

### 2010

• Mikhail Prokopenko, Nihat Ay, Oliver Obst, and Daniel Polani. Phase transitions in least-effort communications. Journal of Statistical Mechanics: Theory and Experiment, 2010(11):P11025, 2010.
[Bibtex] [Abstract] [Details]
This is an author-created, un-copyedited version of an article accepted for publication in J. Stat. Mech. IOP Publishing Ltd is not responsible for any errors or omissions in this version of the manuscript or any version derived from it. The definitive publisher authenticated version is available online at doi: 10.1088/1742-5468/2010/11/P11025.

We critically examine a model that attempts to explain the emergence of power laws (e.g., Zipf's law) in human language. The model is based on the principle of least effort in communications---specifically, the overall effort is balanced between the speaker effort and listener effort, with some trade-off. It has been shown that an information-theoretic interpretation of this principle is sufficiently rich to explain the emergence of Zipf's law in the vicinity of the transition between referentially useless systems (one signal for all referable objects) and indexical reference systems (one signal per object). The phase transition is defined in the space of communication accuracy (information content) expressed in terms of the trade-off parameter. Our study explicitly solves the continuous optimization problem, subsuming a recent, more specific result obtained within a discrete space. The obtained results contrast Zipf's law found by heuristic search (that attained only local minima) in the vicinity of the transition between referentially useless systems and indexical reference systems, with an inverse-factorial (sub-logarithmic) law found at the transition that corresponds to global minima. The inverse-factorial law is observed to be the most representative frequency distribution among optimal solutions.

@Article{ PAOP10,
abstract  = {We critically examine a model that attempts to explain the emergence of power laws (e.g., Zipf's law) in human language. The model is based on the principle of least effort in communications---specifically, the overall effort is balanced between the speaker effort and listener effort,
with some trade-off. It has been shown that an information-theoretic interpretation of this principle is sufficiently rich to explain the emergence of Zipf's law in the vicinity of the transition between referentially useless systems (one signal for all referable objects) and indexical reference systems (one signal per object). The phase transition is defined in the space of communication accuracy (information content) expressed in terms of the trade-off parameter. Our study explicitly solves the continuous optimization problem, subsuming a recent, more specific result obtained within a discrete space. The obtained results contrast Zipf's law found by heuristic search (that attained only local minima) in the vicinity of the transition between referentially useless systems and indexical reference systems, with an inverse-factorial (sub-logarithmic) law found at the transition that corresponds to global minima. The inverse-factorial law is observed to be the most representative frequency distribution among optimal solutions.},
author  = {Mikhail Prokopenko and Nihat Ay and Oliver Obst and Daniel Polani},
journal  = {Journal of Statistical Mechanics: Theory and Experiment},
number  = {11},
pages = {P11025},
title = {Phase transitions in least-effort communications},
url = {http://stacks.iop.org/1742-5468/2010/i=11/a=P11025},
volume  = {2010},
wwwnote  = {<small>This is an author-created, un-copyedited version of an article accepted for publication in J. Stat. Mech. IOP Publishing Ltd is not responsible for any errors or omissions in this version of the manuscript or any version derived from it. The definitive publisher authenticated version is available online at doi: <a href="http://dx.doi.org/10.1088/1742-5468/2010/11/P11025">10.1088/1742-5468/2010/11/P11025</a>.</small>}
,
year = {2010}
}

• Oliver Obst, Joschka Boedecker, and Minoru Asada. Improving Recurrent Neural Network Performance using Transfer Entropy. In Kok Wai Wong, Sumudu B. U. Mendis, and Abdesselam Bouzerdoum, editors, Neural Information Processing. Models and Applications, volume 6444 of Lecture Notes in Computer Science, pages 193-200. Springer, 2010.
[Bibtex] [Abstract] [Details]
Reservoir computing approaches have been successfully ap- plied to a variety of tasks. An inherent problem of these approaches, is, however, their variation in performance due to fixed random initialisa- tion of the reservoir. Self-organised approaches like intrinsic plasticity have been applied to improve reservoir quality, but do not take the task of the system into account. We present an approach to improve the hidden layer of recurrent neural networks, guided by the learning goal of the system. Our reservoir adaptation optimises the information transfer at each individual unit, dependent on properties of the information transfer between input and output of the system. Using synthetic data, we show that this reservoir adaptation improves the performance of offline echo state learning and Recursive Least Squares Online Learning.

@InCollection{ OBA10,
abstract  = {Reservoir computing approaches have been successfully ap- plied to a variety of tasks. An inherent problem of these approaches, is, however, their variation in performance due to fixed random initialisa- tion of the reservoir. Self-organised approaches like intrinsic plasticity have been applied to improve reservoir quality, but do not take the task of the system into account. We present an approach to improve the hidden layer of recurrent neural networks,
guided by the learning goal of the system. Our reservoir adaptation optimises the information transfer at each individual unit, dependent on properties of the information transfer between input and output of the system. Using synthetic data, we show that this reservoir adaptation improves the performance of offline echo state learning and Recursive Least Squares Online Learning. },
author  = {Oliver Obst and Joschka Boedecker and Minoru Asada},
booktitle  = {Neural Information Processing. Models and Applications},
doi = {http://dx.doi.org/10.1007/978-3-642-17534-3_24},
editor  = {Wong, Kok Wai and Mendis, B. Sumudu U. and Bouzerdoum,
Abdesselam},
keywords  = {Information Transfer, echo state networks, recurrent neural networks},
pages = {193--200},
publisher  = {Springer},
series  = {Lecture Notes in Computer Science},
title = {Improving Recurrent Neural Network Performance using Transfer Entropy},
volume  = {6444},
year = {2010}
}

• Michael N. Mayer, Oliver Obst, and Chang Yu-Chen. Time Series Causality Inference using Echo State Networks. In Vincent Vigneron, Vicente Zarzoso, Eric Moreau, Rémi Gribonval, and Emmanuel Vincent, editors, Ninth International Conference on Latent Variable Analysis and Signal Separation, volume 6365 of Lecture Notes in Computer Science, pages 279-286. Springer, Berlin, Heidelberg, 2010.
[Bibtex] [Abstract] [Details]
One potential strength of recurrent neural networks (RNNs) is their -- theoretical -- ability to find a connection between cause and consequence in time series in an constraint-free manner, that is without the use of explicit probability theory. In this work we present a solution which uses the echo state approach for this purpose. Our approach learns probabilities explicitly using an online learning procedure and echo state networks. We also demonstrate the approach using a test model.

@InCollection{ MOY10,
abstract  = {One potential strength of recurrent neural networks (RNNs) is their -- theoretical -- ability to find a connection between cause and consequence in time series in an constraint-free manner, that is without the use of explicit probability theory. In this work we present a solution which uses the echo state approach for this purpose. Our approach learns probabilities explicitly using an online learning procedure and echo state networks. We also demonstrate the approach using a test model.},
author  = {N. Michael Mayer and Oliver Obst and Chang Yu-Chen},
booktitle  = {Ninth International Conference on Latent Variable Analysis and Signal Separation},
doi = {http://dx.doi.org/10.1007/978-3-642-15995-4_35},
editor  = {Vigneron, Vincent and Zarzoso, Vicente and Moreau, Eric and Gribonval, R{\'e}mi and Vincent, Emmanuel},
pages = {279--286},
publisher  = {Springer},
series  = {Lecture Notes in Computer Science},
title = {Time Series Causality Inference using Echo State Networks},
volume  = {6365},
year = {2010}
}

### 2009

• Oliver Obst. Distributed Backpropagation-Decorrelation Learning. In NIPS Workshop: Large-Scale Machine Learning: Parallelism and Massive Datasets, 2009.
[Bibtex] [Abstract] [Details]
In long-term deployments of sensor networks, monitoring the quality of gathered data is a critical issue. Over the time of deployment, exposure to harsh condition may cause sensors to degrade or to fail. If such a degradation remains undetected, the usefulness of a sensor network is greatly reduced. We introduce SODBPDC, a distributed recurrent network architecture, and a method to learn spatio-temporal correlations between different sensors for fault detection in a distributed way. Our approach is evaluated using real sensor network data, and proves to work well with less-than-perfect link qualities and more than 50% of failed sensors.

@InProceedings{ Obst09d,
abstract  = {In long-term deployments of sensor networks, monitoring the quality of gathered data is a critical issue. Over the time of deployment, exposure to harsh condition may cause sensors to degrade or to fail. If such a degradation remains undetected, the usefulness of a sensor network is greatly reduced. We introduce SODBPDC, a distributed recurrent network architecture, and a method to learn spatio-temporal correlations between different sensors for fault detection in a distributed way. Our approach is evaluated using real sensor network data, and proves to work well with less-than-perfect link qualities and more than 50% of failed sensors. },
author  = {Oliver Obst},
booktitle  = {NIPS Workshop: Large-Scale Machine Learning: Parallelism and Massive Datasets},
title = {Distributed Backpropagation-Decorrelation Learning},
year = {2009}
}

• Joschka Boedecker, Oliver Obst, Norbert Michael Mayer, and Minoru Asada. Initialization and Self-Organized Optimization of Recurrent Neural Network Connectivity. HFSP Journal, 3(5):340-349, 2009.
[Bibtex] [Abstract] [Details]
Reservoir computing (RC) is a recent paradigm in the field of recurrent neural networks. Networks in RC have a sparsely and randomly connected fixed hidden layer, and only output connections are trained. RC Networks have recently received increased attention as a mathematical model for generic neural microcircuits, to investigate and explain computations in neocortical columns. Applied to specific tasks, their fixed random connectivity, however, leads to significant variation in performance. Few problem specific optimization procedures are known, which would be important for engineering applications, but also in order to understand how networks in biology are shaped to be optimally adapted to requirements of their environment. We study a general network initialization method using permutation matrices and derive a new unsupervised learning rule based on intrinsic plasticity (IP). The IP based learning uses only local learning, and its aim is to improve network performance in a self-organized way. Using three different benchmarks, we show that networks with permutation matrices for the reservoir connectivity have much more persistent memory than the other methods, but are also able to perform highly non-linear mappings. We also show that IP based on sigmoid transfer functions is limited concerning the output distributions that can be achieved.

@Article{ BOMA09b,
abstract  = {Reservoir computing (RC) is a recent paradigm in the field of recurrent neural networks. Networks in RC have a sparsely and randomly connected fixed hidden layer, and only output connections are trained. RC Networks have recently received increased attention as a mathematical model for generic neural microcircuits, to investigate and explain computations in neocortical columns. Applied to specific tasks, their fixed random connectivity, however,
leads to significant variation in performance. Few problem specific optimization procedures are known, which would be important for engineering applications, but also in order to understand how networks in biology are shaped to be optimally adapted to requirements of their environment. We study a general network initialization method using permutation matrices and derive a new unsupervised learning rule based on intrinsic plasticity (IP). The IP based learning uses only local learning, and its aim is to improve network performance in a self-organized way. Using three different benchmarks, we show that networks with permutation matrices for the reservoir connectivity have much more persistent memory than the other methods, but are also able to perform highly non-linear mappings. We also show that IP based on sigmoid transfer functions is limited concerning the output distributions that can be achieved. },
author  = {Joschka Boedecker and Oliver Obst and Norbert Michael Mayer and Minoru Asada},
doi = {10.2976/1.3240502},
journal  = {HFSP Journal},
keywords  = {recurrent neural networks, intrinsic plasticity, reservoir computing},
month = oct,
number  = {5},
pages = {340--349},
title = {Initialization and Self-Organized Optimization of Recurrent Neural Network Connectivity},
url = {http://dx.doi.org/10.2976/1.3240502},
volume  = {3},
year = {2009}
}

• Oliver Obst. Construction and training of a recurrent neural network. Patent application WO/2010/144947, 2009.
[Bibtex] [Details]
@Misc{ Obst09c,
author  = {Oliver Obst},
howpublished  = {Patent application WO/2010/144947},
keywords  = {anomaly, fault detection, recurrent neural networks},
month = jun,
title = {Construction and training of a recurrent neural network},
year = {2009}
}

• Oliver Obst. Distributed Fault Detection using a Recurrent Neural Network. In Proceedings of the International Conference on Information Processing in Sensor Networks (IPSN 2009), pages 373-374. IEEE Computer Society, 2009.
[Bibtex] [Details]
@InProceedings{ Obst09a,
author  = {Oliver Obst},
booktitle  = {Proceedings of the International Conference on Information Processing in Sensor Networks (IPSN 2009)},
isbn = {978-1-60558-371-6},
keywords  = {anomaly, fault detection, echo state networks, recurrent neural networks, sensor networks},
month = apr,
pages = {373--374},
publisher  = {IEEE Computer Society},
title = {Distributed Fault Detection using a Recurrent Neural Network},
year = {2009}
}

• Joschka Boedecker, Oliver Obst, Norbert Michael Mayer, and Minoru Asada. Studies on Reservoir Initialization and Dynamics Shaping in Echo State Networks. In Michel Verleysen, editor, Proceedings of the 17th European Symposium On Artificial Neural Networks (ESANN'09), pages 227-232, Evere, Belgium, apr 2009. D-Side Publications.
[Bibtex] [Abstract] [Details]
The fixed random connectivity of networks in reservoir computing leads to significant variation in performance. Only few problem specific optimization procedures are known to date. We study a general initialization method using permutation matrices and derive a new unsupervised learning rule based on intrinsic plasticity (IP) for echo state networks. Using three different benchmarks, we show that networks with permutation matrices for the reservoir connectivity have much longer memory than the other methods, but are also able to perform highly non-linear mappings. We also show that IP based on sigmoid transfer functions is limited concerning the output distributions that can be achieved.

@InProceedings{ BOMA09a,
abstract  = {The fixed random connectivity of networks in reservoir computing leads to significant variation in performance. Only few problem specific optimization procedures are known to date. We study a general initialization method using permutation matrices and derive a new unsupervised learning rule based on intrinsic plasticity (IP) for echo state networks. Using three different benchmarks, we show that networks with permutation matrices for the reservoir connectivity have much longer memory than the other methods, but are also able to perform highly non-linear mappings. We also show that IP based on sigmoid transfer functions is limited concerning the output distributions that can be achieved. },
author  = {Joschka Boedecker and Oliver Obst and Norbert Michael Mayer and Minoru Asada},
booktitle  = {Proceedings of the 17th European Symposium On Artificial Neural Networks ({ESANN}'09)},
editor  = {Michel Verleysen},
month = apr,
pages = {227--232},
publisher  = {D-Side Publications},
title = {Studies on Reservoir Initialization and Dynamics Shaping in Echo State Networks},
year = {2009}
}

### 2008

• Rosalind X. Wang, Joseph T. Lizier, Oliver Obst, Mikhail Prokopenko, and Peter Wang. Spatiotemporal Anomaly Detection in Gas Monitoring Sensor Networks. In Roberto Verdone, editor, Wireless Sensor Networks, volume 4913 of Lecture Notes in Computer Science, pages 90-105. Springer, Berlin, Heidelberg, feb 2008.
[Bibtex] [Abstract] [Details]
In this paper, we use Bayesian Networks as a means for unsupervised learning and anomaly (event) detection in gas monitoring sensor networks for underground coal mines. We show that the Bayesian Network model can learn cyclical baselines for gas concentrations, thus reducing false alarms usually caused by flatline thresholds. Further, we show that the system can learn dependencies between changes of concentration in different gases and at multiple locations. We define and identify new types of events that can occur in a sensor network. In particular, we analyse joint events in a group of sensors based on learning the Bayesian model of the system, contrasting these events with merely aggregating single events. We demonstrate that anomalous events in individual gas data might be explained if considered jointly with the changes in other gases. Vice versa, a network-wide spatiotemporal anomaly may be detected even if individual sensor readings were within their thresholds. The presented Bayesian approach to spatiotemporal anomaly detection is applicable to a wide range of sensor networks.

@InCollection{ WLO+08,
abstract  = {In this paper, we use Bayesian Networks as a means for unsupervised learning and anomaly (event) detection in gas monitoring sensor networks for underground coal mines. We show that the Bayesian Network model can learn cyclical baselines for gas concentrations, thus reducing false alarms usually caused by flatline thresholds. Further, we show that the system can learn dependencies between changes of concentration in different gases and at multiple locations. We define and identify new types of events that can occur in a sensor network. In particular, we analyse joint events in a group of sensors based on learning the Bayesian model of the system, contrasting these events with merely aggregating single events. We demonstrate that anomalous events in individual gas data might be explained if considered jointly with the changes in other gases. Vice versa, a network-wide spatiotemporal anomaly may be detected even if individual sensor readings were within their thresholds. The presented Bayesian approach to spatiotemporal anomaly detection is applicable to a wide range of sensor networks. },
author  = {X. Rosalind Wang and Joseph T. Lizier and Oliver Obst and Mikhail Prokopenko and Peter Wang},
booktitle  = {Wireless Sensor Networks},
doi = {10.1007/978-3-540-77690-1},
editor  = {Roberto Verdone},
month = feb,
pages = {90--105},
publisher  = {Springer},
series  = {Lecture Notes in Computer Science},
title = {Spatiotemporal Anomaly Detection in Gas Monitoring Sensor Networks},
volume  = {4913},
year = {2008}
}

• Oliver Obst, Rosalind X. Wang, and Mikhail Prokopenko. Using Echo State Networks for Anomaly Detection in Underground Coal Mines. In Proceedings of the International Conference on Information Processing in Sensor Networks (IPSN 2008), pages 219-229. IEEE Computer Society, apr 2008.
[Bibtex] [Abstract] [Details]
We investigate the problem of identifying anomalies in monitoring critical gas concentrations using a sensor network in an underground coal mine. In this domain, one of the main problems is a provision of mine specific anomaly detection, with cyclical (moving) instead of flatline (static) alarm threshold levels. An additional practical difficulty in modelling a specific mine is the lack of fully labelled data of normal and abnormal situations. We present an approach addressing these difficulties based on echo state networks learning mine specific anomalies when only normal data is available. Echo state networks utilize incremental updates driven by new sensor readings, thus enabling a detection of anomalies at any time during the sensor network operation. We evaluate this approach against a benchmark -- Bayes Network based anomaly detection, and observe that the quality of the overall predictions is comparable to the benchmark. However, the echo state networks maintain the same level of predictive accuracy for data from multiple sources. Therefore, the ability of echo state networks to model dynamical systems make this approach more suitable for anomaly detection and predictions in sensor networks.

@InProceedings{ OWP08,
abstract  = {We investigate the problem of identifying anomalies in monitoring critical gas concentrations using a sensor network in an underground coal mine. In this domain, one of the main problems is a provision of mine specific anomaly detection, with cyclical (moving) instead of flatline (static) alarm threshold levels. An additional practical difficulty in modelling a specific mine is the lack of fully labelled data of normal and abnormal situations. We present an approach addressing these difficulties based on echo state networks learning mine specific anomalies when only normal data is available. Echo state networks utilize incremental updates driven by new sensor readings, thus enabling a detection of anomalies at any time during the sensor network operation. We evaluate this approach against a benchmark -- Bayes Network based anomaly detection, and observe that the quality of the overall predictions is comparable to the benchmark. However, the echo state networks maintain the same level of predictive accuracy for data from multiple sources. Therefore, the ability of echo state networks to model dynamical systems make this approach more suitable for anomaly detection and predictions in sensor networks. },
author  = {Oliver Obst and X. Rosalind Wang and Mikhail Prokopenko},
booktitle  = {Proceedings of the International Conference on Information Processing in Sensor Networks (IPSN 2008)},
isbn = {978-0-7695-3157-1},
keywords  = {echo state networks, recurrent neural networks, sensor networks, bayesian networks, anomaly, novelty, coal mine},
month = apr,
pages = {219--229},
publisher  = {IEEE Computer Society},
title = {Using Echo State Networks for Anomaly Detection in Underground Coal Mines},
year = {2008}
}

• Frank Dylla, Alexander Ferrein, Gerhard Lakemeyer, Jan Murray, Oliver Obst, Thomas Röfer, Stefan Schiffer, Frieder Stolzenburg, Ubbo Visser, and Thomas Wagner. Computers in Sport, chapter Approaching a Formal Soccer Theory from the Behavior Specification in Robotic Soccer, pages 161-186. Bioengineering. WIT Press, 2008. ISBN 978-1-84564-064-4.
[Bibtex] [Abstract] [Details]
You can order the book at Amazon.

This chapter discusses a top-down approach to modelling soccer knowledge, as it can be found in soccer theory books. The goal is to model soccer strategies and tactics in a way that they are usable for multiple robotic soccer leagues in the RoboCup. We investigate if and how soccer theory can be formalized such that specification and execution are possible. The advantage is clear: theory abstracts from hardware and from specifi c situations in different leagues. We introduce basic primitives compliant with the terminology known in soccer theory, discuss an example on an abstract level and formalize it. The formalization of soccer presented here is appealing. It goes beyond the behaviour specifi cation of soccer playing robots. For sports science a unified formal soccer theory might help to better understand and to formulate basic concepts in soccer. The possibility of the formalization to develop computer programs, which allow to simulate and to reason about soccer moves, might also take sports science a step further.

@InBook{ DFL+08,
abstract  = {This chapter discusses a top-down approach to modelling soccer knowledge, as it can be found in soccer theory books. The goal is to model soccer strategies and tactics in a way that they are usable for multiple robotic soccer leagues in the RoboCup. We investigate if and how soccer theory can be formalized such that specification and execution are possible. The advantage is clear: theory abstracts from hardware and from specifi c situations in different leagues. We introduce basic primitives compliant with the terminology known in soccer theory, discuss an example on an abstract level and formalize it. The formalization of soccer presented here is appealing. It goes beyond the behaviour specifi cation of soccer playing robots. For sports science a unified formal soccer theory might help to better understand and to formulate basic concepts in soccer. The possibility of the formalization to develop computer programs, which allow to simulate and to reason about soccer moves, might also take sports science a step further. },
author  = {Frank Dylla and Alexander Ferrein and Gerhard Lakemeyer and Jan Murray and Oliver Obst and Thomas R{\"o}fer and Stefan Schiffer and Frieder Stolzenburg and Ubbo Visser and Thomas Wagner},
chapter  = {Approaching a Formal Soccer Theory from the Behavior Specification in Robotic Soccer},
editor  = {Peter Dabnicki and Arnold Baca},
keywords  = {robocup, robotic soccer, formal methods, spatial reasoning, simulation},
note = {ISBN 978-1-84564-064-4.},
pages = {161--186},
publisher  = {WIT Press},
series  = {Bioengineering},
title = {Computers in Sport},
wwwnote  = {You can <a href="http://www.amazon.com/gp/product/1845640640?ie=UTF8&tag=droliobs-20&linkCode=as2&camp=1789&creative=9325&creativeASIN=1845640640">order the book at Amazon</a>.},
year = {2008}
}

### 2007

• Oliver Obst. Controlling Physical Multiagent Teams: Getting League-Independent Results from RoboCup Soccer. Number 304 in DISKI -- Dissertations in Artificial Intelligence. Aka / IOS Press, 2007. ISBN 978-1-58603-705-5.
[Bibtex] [Details]
You can buy the book at Amazon.
@Book{ Obst07,
author  = {Oliver Obst},
note = {ISBN 978-1-58603-705-5.},
number  = {304},
publisher  = {Aka / IOS Press},
series  = {DISKI -- Dissertations in Artificial Intelligence},
title = {Controlling Physical Multiagent Teams: Getting League-Independent Results from RoboCup Soccer},
url = {http://tinyurl.com/24vxlh/},
wwwnote  = {You can <a href="http://tinyurl.com/24vxlh">buy the book at Amazon</a>.},
year = {2007}
}

• Norbert Michael Mayer, Joschka Boedecker, Rodrigo da Silva Guerra, Oliver Obst, and Minoru Asada. 3D2Real: Simulation League Finals in Real Robots. In Gerhard Lakemeyer, Elizabeth Sklar, Domenico G. Sorrenti, and Tomoichi Takahashi, editors, RoboCup 2006: Robot Soccer World Cup~X, volume 4434 of Lecture Notes in Artificial Intelligence, pages 25-34, Berlin, Heidelberg, 2007. Springer.
[Bibtex] [Abstract] [Details]
We present a road map for a joint project of the simulation league and the humanoid league that we call 3D2Real. This project is concerned with the integration of these two leagues which is becoming increasingly important as the research fields are converging. Currently, a lot of work is duplicated across the leagues, collaboration is sparse, and knowhow is not transfered effectively. This binds resources to solve the same problems over and over again. To address this, we discuss the current situation of both leagues with respect to these points and focus on open issues that have to be fixed. In addition, we describe existing open standards and contributions from the RoboCup community that we plan to use for the project. As a milestone, we propose to conduct the finals of the 3D simulation tournament on real robots by the year 2008. Finally, we propose a database of simulated parts and algorithms in which each league can benefit and contribute with their expertise. These contributions facilitate synergies to be used across individual leagues for the benefit of the RoboCup project and the year 2050 goal.

@InProceedings{ MBG+07,
abstract  = {We present a road map for a joint project of the simulation league and the humanoid league that we call 3D2Real. This project is concerned with the integration of these two leagues which is becoming increasingly important as the research fields are converging. Currently, a lot of work is duplicated across the leagues, collaboration is sparse, and knowhow is not transfered effectively. This binds resources to solve the same problems over and over again. To address this, we discuss the current situation of both leagues with respect to these points and focus on open issues that have to be fixed. In addition, we describe existing open standards and contributions from the RoboCup community that we plan to use for the project. As a milestone, we propose to conduct the finals of the 3D simulation tournament on real robots by the year 2008. Finally, we propose a database of simulated parts and algorithms in which each league can benefit and contribute with their expertise. These contributions facilitate synergies to be used across individual leagues for the benefit of the RoboCup project and the year 2050 goal.},
author  = {Norbert Michael Mayer and Joschka Boedecker and Rodrigo da Silva Guerra and Oliver Obst and Minoru Asada},
booktitle  = {RoboCup 2006: Robot Soccer World Cup~X},
doi = {10.1007/978-3-540-74024-7},
editor  = {Gerhard Lakemeyer and Elizabeth Sklar and Domenico G. Sorrenti and Tomoichi Takahashi},
keywords  = {robocup, simulation, humanoid, robotic soccer, road map},
pages = {25--34},
publisher  = {Springer},
series  = {Lecture Notes in Artificial Intelligence},
title = {{3D2Real}: Simulation League Finals in Real Robots},
volume  = {4434},
year = {2007}
}

• Michael Quinlan, Oliver Obst, and Stephan Chalup. Towards Autonomous Strategy Decisions in the RoboCup Four-Legged League. In Proceedings of the Seventh IJCAI International Workshop on Nonmontonic Reasoning, Action and Change, 2007.
[Bibtex] [Abstract] [Details]
Each of the soccer leagues at RoboCup addresses a different aspect of the complex soccer task. The Four-Legged League is the only robot soccer league where more than two real legged robots play in a team. High levels of noise hamper vision and localisation, and therefore deliberate passing occurs rarely in normal game play. The present study highlights some connections to the simulation leagues where eleven agents play in a team and successful passes occur frequently. In the simulation leagues the development of successful global team strategies is at the centre of interest. The experiments in the present study evaluated the impact of varying global team strategies in the Four-Legged League. The NUbots 2006 system was tested against more aggressive and more defensive strategies. The results indicate that global team tactics should be considered in conjunction with a team's style of play. A set of metrics was developed which may enable a future robot soccer team to observe, reason, and modify its global strategy to suit that of an opposing team.

@InProceedings{ QOC07,
abstract  = {Each of the soccer leagues at RoboCup addresses a different aspect of the complex soccer task. The Four-Legged League is the only robot soccer league where more than two real legged robots play in a team. High levels of noise hamper vision and localisation, and therefore deliberate passing occurs rarely in normal game play. The present study highlights some connections to the simulation leagues where eleven agents play in a team and successful passes occur frequently. In the simulation leagues the development of successful global team strategies is at the centre of interest. The experiments in the present study evaluated the impact of varying global team strategies in the Four-Legged League. The NUbots 2006 system was tested against more aggressive and more defensive strategies. The results indicate that global team tactics should be considered in conjunction with a team's style of play. A set of metrics was developed which may enable a future robot soccer team to observe, reason, and modify its global strategy to suit that of an opposing team. },
author  = {Michael Quinlan and Oliver Obst and Stephan Chalup},
booktitle  = {Proceedings of the Seventh IJCAI International Workshop on Nonmontonic Reasoning, Action and Change},
title = {Towards Autonomous Strategy Decisions in the {RoboCup}
Four-Legged League},
year = {2007}
}

### 2006

• Oliver Obst and Joschka Boedecker. Flexible Coordination Of Multiagent Team Behavior Using HTN Planning. In Itsuki Noda, Adam Jacoff, Ansgar Bredenfeld, and Yasutake Takahashi, editors, RoboCup 2005: Robot Soccer World Cup~IX, Lecture Notes in Artificial Intelligence, pages 521-528. Springer, Berlin, Heidelberg, New York, 2006.
[Bibtex] [Abstract] [Details]
The domain of robotic soccer is known as a highly dynamic and non-deterministic environment for multiagent research. We introduce an approach using Hierarchical Task Network planning in each of the agents for high-level coordination and description of team strategies. Our approach facilitates the maintenance of expert knowledge specified as team strategies separated from the agent implementation. By combining high level plans with reactive basic operators, agents can pursue a grand strategy while staying reactive to changes in the environment. Our results show that the use of a planner in a multiagent system is both possible and useful despite the constraints in dynamic environments.

@InCollection{ OB06,
abstract  = {The domain of robotic soccer is known as a highly dynamic and non-deterministic environment for multiagent research. We introduce an approach using Hierarchical Task Network planning in each of the agents for high-level coordination and description of team strategies. Our approach facilitates the maintenance of expert knowledge specified as team strategies separated from the agent implementation. By combining high level plans with reactive basic operators, agents can pursue a grand strategy while staying reactive to changes in the environment. Our results show that the use of a planner in a multiagent system is both possible and useful despite the constraints in dynamic environments. },
address  = {Berlin, Heidelberg, New York},
author  = {Oliver Obst and Joschka Boedecker},
booktitle  = {{R}obo{C}up 2005: Robot Soccer World Cup~{IX}},
editor  = {Itsuki Noda and Adam Jacoff and Ansgar Bredenfeld and Yasutake Takahashi},
pages = {521--528},
publisher  = {Springer},
series  = {Lecture Notes in Artificial Intelligence},
title = {Flexible Coordination Of Multiagent Team Behavior Using {HTN} Planning},
year = {2006}
}

• Heni Ben Amor, Jan Murray, and Oliver Obst. AI Game Programming Wisdom 3, chapter Fast, Neat and Under Control: Inverse Steering Behaviors for Physical Autonomous Agents, pages 221-232. Charles River Media, Boston, MA, 2006. ISBN 1-58450-457-9
[Bibtex] [Abstract] [Details]
You can get the whole book at Amazon.

Steering behaviors are a set of reactive algorithms used for navigating autonomous agents in their environment. Combinations of steering behaviors can be used to create complex, interesting and lifelike movement. However, special care has to be given to their arbitration. If done the wrong way, the arbitration can lead to suboptimal, undesired, or even catastrophic results in certain situations. This article presents a solution to these problems by introducing inverse steering behaviors (ISBs) for controlling physical agents. Inverse steering behaviors change the original concept of steering behaviors and facilitate improved arbitration between different options by using cost based heuristics. The approach is demonstrated on a soccer playing agent.

@InBook{ AMO06,
abstract  = {Steering behaviors are a set of reactive algorithms used for navigating autonomous agents in their environment. Combinations of steering behaviors can be used to create complex, interesting and lifelike movement. However,
special care has to be given to their arbitration. If done the wrong way, the arbitration can lead to suboptimal,
undesired, or even catastrophic results in certain situations. This article presents a solution to these problems by introducing inverse steering behaviors (ISBs) for controlling physical agents. Inverse steering behaviors change the original concept of steering behaviors and facilitate improved arbitration between different options by using cost based heuristics. The approach is demonstrated on a soccer playing agent.},
author  = {Heni {Ben Amor} and Jan Murray and Oliver Obst},
chapter  = {Fast, Neat and Under Control: Inverse Steering Behaviors for Physical Autonomous Agents},
editor  = {Steve Rabin},
note = {ISBN 1-58450-457-9},
pages = {221--232},
publisher  = {Charles River Media},
title = {AI Game Programming Wisdom 3},
wwwnote  = {You can <a href="http://www.amazon.com/gp/product/1584504579?ie=UTF8&tag=droliobs-20&linkCode=as2&camp=1789&creative=9325&creativeASIN=1584504579">get the whole book at Amazon</a>.},
year = {2006}
}

• Oliver Obst. Using a Planner for Coordination Of Multiagent Team Behavior. In Rafael H. Bordini, Mehdi Dastani, Jürgen Dix, and Amal ElFallah Seghrouchni, editors, Programming Multi-Agent Systems: Third International Workshop, ProMAS 2005, Utrecht, The Netherlands, July 26, 2005, Revised and Invited Papers, volume 3862 of Lecture Notes in Computer Science, pages 90-100. Springer, Berlin, 2006.
[Bibtex] [Abstract] [Details]
We present an approach to coordinate the behavior of a multiagent team using an HTN planning procedure. To coordinate teams, high level tasks have to be broken down into subtasks which is a basic operation in HTN planners. We are using planners in each of the agents to incorporate domain knowledge and to make agents follow a specified team strategy. With our approach, agents coordinate deliberatively and still maintain a high degree of reactivity. In our implementation for use in RoboCup Simulation League, first results were already very promising. Using a planner leads to better separation of agent code and expert knowledge.

@InCollection{ Obst06,
abstract  = {We present an approach to coordinate the behavior of a multiagent team using an HTN planning procedure. To coordinate teams, high level tasks have to be broken down into subtasks which is a basic operation in HTN planners. We are using planners in each of the agents to incorporate domain knowledge and to make agents follow a specified team strategy. With our approach, agents coordinate deliberatively and still maintain a high degree of reactivity. In our implementation for use in RoboCup Simulation League, first results were already very promising. Using a planner leads to better separation of agent code and expert knowledge. },
annote  = {ISBN: 3-540-32616-2},
author  = {Oliver Obst},
booktitle  = {Programming Multi-Agent Systems: Third International Workshop, ProMAS 2005, Utrecht, The Netherlands, July 26,
2005, Revised and Invited Papers},
editor  = {Rafael H. Bordini and Mehdi Dastani and J{\"u}rgen Dix and Amal {ElFallah Seghrouchni}},
html = {http://dx.doi.org/10.1007/11678823_6},
month = mar,
pages = {90 -- 100},
publisher  = {Springer},
series  = {Lecture Notes in Computer Science},
title = {Using a Planner for Coordination Of Multiagent Team Behavior},
volume  = {3862},
year = {2006}
}

### 2005

• Frank Dylla, Alexander Ferrein, Gerhardt Lakemeyer, Jan Murray, Oliver Obst, Thomas Röfer, Frieder Stolzenburg, Ubbo Visser, and Thomas Wagner. Towards a League-Independent Qualitative Soccer Theory for RoboCup. In Daniele Nardi, Martin Riedmiller, Claude Sammut, and José Santos-Victor, editors, RoboCup 2004: Robot Soccer World Cup VIII, volume 3276 of Lecture Notes in Artificial Intelligence, pages 611-618. Springer, Berlin, Heidelberg, New York, 2005.
[Bibtex] [Abstract] [Details]
A preliminary version appeared as Fachberichte Informatik 6/2004, Universität Koblenz-Landau.

The paper discusses a top-down approach to model soccer knowledge, as it can be found in soccer theory books. The goal is to model soccer strategies and tactics in a way that they are usable for multiple RoboCup soccer leagues, i.e. for different hardware platforms. We investigate if and how soccer theory can be formalized such that specification and execution is possible. The advantage is clear: theory abstracts from hardware and from specific situations in different leagues. Such a qualitative abstraction is well suited for comparing and evaluating different systems and approaches. We introduce basic primitives compliant with the terminology known in soccer theory, discuss an example on an abstract level and formalize it. We then consider aspects of different RoboCup leagues in a case study and examine how examples can be instantiated in three different leagues.

@InCollection{ DFL+05,
abstract  = {The paper discusses a top-down approach to model soccer knowledge, as it can be found in soccer theory books. The goal is to model soccer strategies and tactics in a way that they are usable for multiple RoboCup soccer leagues,
i.e. for different hardware platforms. We investigate if and how soccer theory can be formalized such that specification and execution is possible. The advantage is clear: theory abstracts from hardware and from specific situations in different leagues. Such a qualitative abstraction is well suited for comparing and evaluating different systems and approaches. We introduce basic primitives compliant with the terminology known in soccer theory, discuss an example on an abstract level and formalize it. We then consider aspects of different RoboCup leagues in a case study and examine how examples can be instantiated in three different leagues.},
address  = {Berlin, Heidelberg, New York},
author  = {Frank Dylla and Alexander Ferrein and Gerhardt Lakemeyer and Jan Murray and Oliver Obst and Thomas R{\"o}fer and Frieder Stolzenburg and Ubbo Visser and Thomas Wagner},
booktitle  = {RoboCup 2004: Robot Soccer World Cup VIII},
editor  = {Daniele Nardi and Martin Riedmiller and Claude Sammut and Jos{\'e} Santos-Victor},
pages = {611--618},
publisher  = {Springer},
series  = {Lecture Notes in Artificial Intelligence},
title = {Towards a {L}eague-{I}ndependent {Q}ualitative {S}occer {T}heory for {R}obo{C}up},
volume  = {3276},
wwwnote  = {A <a href="http://www.uni-koblenz.de/fb4/publikationen/gelbereihe/RR-6-2004.pdf">preliminary&nbsp;version</a> appeared as Fachberichte Informatik 6/2004, Universit{\"a}t Koblenz-Landau.},
year = {2005}
}

• Oliver Obst and Markus Rollmann. SPARK -- A Generic Simulator for Physical Multiagent Simulations. Computer Systems Science and Engineering, 20(5):347-356, sep 2005.
[Bibtex] [Abstract] [Details]
We describe a new multi-agent simulation system, called Spark, for agents in three-dimensional environments. Our goal in creating Spark was to provide a high degree of flexibility for creating new types of simulations. We implemented a flexible application framework and exhausted the idea of replaceable components in the resulting system. In comparison to specialized simulators, users can effortlessly create new simulations by using a scene description language. Spark is a powerful tool to state different multi-agent research questions. It was already used as official simulator for the first three-dimensional RoboCup Simulation League competition.

@Article{ OR05,
abstract  = {We describe a new multi-agent simulation system, called Spark, for agents in three-dimensional environments. Our goal in creating Spark was to provide a high degree of flexibility for creating new types of simulations. We implemented a flexible application framework and exhausted the idea of replaceable components in the resulting system. In comparison to specialized simulators, users can effortlessly create new simulations by using a scene description language. Spark is a powerful tool to state different multi-agent research questions. It was already used as official simulator for the first three-dimensional RoboCup Simulation League competition. },
author  = {Oliver Obst and Markus Rollmann},
journal  = {Computer Systems Science and Engineering},
month = sep,
number  = {5},
pages = {347--356},
title = {{SPARK} -- {A} {G}eneric {S}imulator for {P}hysical {M}ultiagent {S}imulations},
volume  = {20},
year = {2005}
}

• Pedro Lima, Lu 'i, Levent Akin, Adam Jacoff, Gerhard Kraezschmar, Beng Kiat Ng, Oliver Obst, Thomas Röfer, Yasutake Takahashi, and Changjiu Zhou. RoboCup 2004 Competitions and Symposium: A Small Kick for Robots, a Giant Score for Science. AI Magazine, 26(2):36-61, Summer 2005.
[Bibtex] [Abstract] [Details]
The linked PDF is a preliminary version.

RoboCup is an international initiative with the main goals of fostering research and education in Artificial Intelligence and Robotics, as well as of promoting Science and Technology to world citizens. The idea is to provide a standard problem where a wide range of technologies can be integrated and examined, as well as being used for project-oriented education, and to organize annual events open to the general public, where different solutions to the problem are compared. The 8th annual of RoboCup -- RoboCup2004 -- was held in Lisbon, Portugal, from 27 June to 5 July. In this paper a general description of RoboCup2004, namely summaries concerning teams, participants, distribution per leagues, main research advances, as well as detailed descriptions for each league, are presented.

@Article{ LCA+05,
abstract  = {RoboCup is an international initiative with the main goals of fostering research and education in Artificial Intelligence and Robotics, as well as of promoting Science and Technology to world citizens. The idea is to provide a standard problem where a wide range of technologies can be integrated and examined, as well as being used for project-oriented education, and to organize annual events open to the general public, where different solutions to the problem are compared. The 8th annual of RoboCup -- RoboCup2004 -- was held in Lisbon, Portugal, from 27 June to 5 July. In this paper a general description of RoboCup2004, namely summaries concerning teams,
participants, distribution per leagues, main research advances, as well as detailed descriptions for each league,
are presented.},
author  = {Pedro Lima and Lu{\'\i}s Cust{\'o}dio and Levent Akin and Adam Jacoff and Gerhard Kraezschmar and Beng Kiat Ng and Oliver Obst and Thomas R{\"o}fer and Yasutake Takahashi and Changjiu Zhou},
journal  = {AI Magazine},
month = {Summer},
number  = {2},
pages = {36--61},
title = {{RoboCup} 2004 Competitions and Symposium: A Small Kick for Robots, a Giant Score for Science},
volume  = {26},
wwwnote  = {The linked PDF is a preliminary version.},
year = {2005}
}

### 2004

• Oliver Obst. Using Model-Based Diagnosis to Build Hypotheses about Spatial Environments. In Daniel Polani, Andrea Bonarini, Brett Browning, and Kazuo Yoshida, editors, RoboCup 2003: Robot Soccer World Cup~VII, Lecture Notes in Artificial Intelligence, pages 518-525. Springer, Berlin, Heidelberg, New York, 2004.
[Bibtex] [Abstract] [Details]
Extended version appeared as Fachberichte Informatik 4/2002, Universität Koblenz-Landau.

We present a method to build a hypothesis on the condition of the environment in which a robotic multi-agent team moves. Initially the robots have a default assumption about the conditions of the floor and on how moving under these condition works. For certain parts of the environment however, the default assumption may be wrong and moving around does not work in the expected way. Now the robotic team builds a hypothesis on the conditions of the yet unvisited part of the environment, so resources can be saved by avoiding areas that possibly also contain obstacles. \par For a description of the environment and of the observations of the robots, we use propositional formulæ in a way similar to computing a diagnosis for electrical circuits. To actually compute the hypothesis, we need to compute models of the given set of clauses, where the extension of the \emph{ab}-literal is minimal. The description of the environment can be generated automatically, and the proposed method is flexible so that different kinds of topologies can be covered.

@InCollection{ Obst04,
abstract  = {We present a method to build a hypothesis on the condition of the environment in which a robotic multi-agent team moves. Initially the robots have a default assumption about the conditions of the floor and on how moving under these condition works. For certain parts of the environment however, the default assumption may be wrong and moving around does not work in the expected way. Now the robotic team builds a hypothesis on the conditions of the yet unvisited part of the environment, so resources can be saved by avoiding areas that possibly also contain obstacles. \par For a description of the environment and of the observations of the robots, we use propositional formul{\ae} in a way similar to computing a diagnosis for electrical circuits. To actually compute the hypothesis, we need to compute models of the given set of clauses, where the extension of the \emph{ab}-literal is minimal. The description of the environment can be generated automatically, and the proposed method is flexible so that different kinds of topologies can be covered.},
address  = {Berlin, Heidelberg, New York},
author  = {Oliver Obst},
booktitle  = {RoboCup 2003: Robot Soccer World Cup~VII},
editor  = {Daniel Polani and Andrea Bonarini and Brett Browning and Kazuo Yoshida},
pages = {518 -- 525},
publisher  = {Springer},
series  = {Lecture Notes in Artificial Intelligence},
title = {Using Model-Based Diagnosis to Build Hypotheses about Spatial Environments},
wwwnote  = {<a href="http://www.uni-koblenz.de/fb4/publikationen/gelbereihe/RR-12-01.pdf">Extended&nbsp;version</a> appeared as Fachberichte Informatik 4/2002, Universit{\"a}t Koblenz-Landau.},
year = {2004}
}

• Marco Kögler and Oliver Obst. Simulation League: The Next Generation. In Daniel Polani, Andrea Bonarini, Brett Browning, and Kazuo Yoshida, editors, RoboCup 2003: Robot Soccer World Cup~VII, volume 3020 of Lecture Notes in Artificial Intelligence, pages 458-469. Springer, Berlin, Heidelberg, New York, 2004.
[Bibtex] [Abstract] [Details]
Extended version appeared as Fachberichte Informatik 3/2003, Universität Koblenz-Landau.

We present a modular approach to model multi-agent simulations in 3D environments. Using this approach, we implemented a generic simulator which is totally decoupled from the actual simulation it performs. We believe that for Soccer Simulation League a transition to 3D states exiting new research problems and equally makes it more attractive to watch for spectators. We are proposing to use our framework as basis for a next generation Soccer Server.

@InCollection{ KO04,
abstract  = { We present a modular approach to model multi-agent simulations in 3D environments. Using this approach, we implemented a generic simulator which is totally decoupled from the actual simulation it performs. We believe that for Soccer Simulation League a transition to 3D states exiting new research problems and equally makes it more attractive to watch for spectators. We are proposing to use our framework as basis for a next generation Soccer Server. },
address  = {Berlin, Heidelberg, New York},
author  = {Marco K{\"o}gler and Oliver Obst},
booktitle  = {RoboCup 2003: Robot Soccer World Cup~VII},
editor  = {Daniel Polani and Andrea Bonarini and Brett Browning and Kazuo Yoshida},
pages = {458 -- 469},
publisher  = {Springer},
series  = {Lecture Notes in Artificial Intelligence},
title = {Simulation League: The Next Generation},
volume  = {3020},
wwwnote  = {<a href="http://www.uni-koblenz.de/fb4/publikationen/gelbereihe/RR-3-2003.pdf">Extended&nbsp;version</a> appeared as Fachberichte Informatik 3/2003, Universit{\"a}t Koblenz-Landau.},
year = {2004}
}

• Oliver Obst and Markus Rollmann. SPARK -- A Generic Simulator for Physical Multiagent Simulations. In Gabriela Lindemann, Jörg Denzinger, Ingo J. Timm, and Rainer Unland, editors, Multiagent System Technologies -- Proceedings of the MATES 2004, volume 3187 of Lecture Notes in Artificial Intelligence, pages 243-257. Springer, sep 2004.
[Bibtex] [Abstract] [Details]
Preliminary version appeared as Fachberichte Informatik 7/2004, Universität Koblenz-Landau.

In this paper we describe a new multi-agent simulation system, called Spark, for physical agents in three-dimensional environments. Our goal in creating Spark was to provide a great amount of flexibility for creating new types of agents and simulations. To achieve this, we implemented a flexible application framework and exhausted the idea of replaceable components in the resulting system. In comparison to specialized simulators, users can effortlessly create new simulations by using a scene description language. Spark is a powerful and flexible tool to state different multi-agent research questions. It is used as official simulator for the first three-dimensional RoboCup Simulation League competition. We present the concepts we used to achieve the flexibility in our system and show how we seamlessly integrated the different subsystems into one user-friendly framework.

@InCollection{ OR04b,
abstract  = { In this paper we describe a new multi-agent simulation system, called Spark, for physical agents in three-dimensional environments. Our goal in creating Spark was to provide a great amount of flexibility for creating new types of agents and simulations. To achieve this, we implemented a flexible application framework and exhausted the idea of replaceable components in the resulting system. In comparison to specialized simulators, users can effortlessly create new simulations by using a scene description language. Spark is a powerful and flexible tool to state different multi-agent research questions. It is used as official simulator for the first three-dimensional RoboCup Simulation League competition. We present the concepts we used to achieve the flexibility in our system and show how we seamlessly integrated the different subsystems into one user-friendly framework.},
author  = {Oliver Obst and Markus Rollmann},
booktitle  = {Multiagent {S}ystem {T}echnologies -- {P}roceedings of the {MATES} 2004},
editor  = {Gabriela Lindemann and J{\"o}rg Denzinger and Ingo J. Timm and Rainer Unland},
month = sep,
pages = {243--257},
publisher  = {Springer},
series  = {Lecture Notes in Artificial Intelligence},
title = {{SPARK} -- {A} {G}eneric {S}imulator for {P}hysical {M}ultiagent {S}imulations},
volume  = {3187},
wwwnote  = {<a href="http://www.uni-koblenz.de/fb4/publikationen/gelbereihe/RR-7-2004.pdf">Preliminary&nbsp;version</a> appeared as Fachberichte Informatik 7/2004, Universit{\"a}t Koblenz-Landau.},
year = {2004}
}

### 2003

• Oliver Obst and Daniel Polani. Simulation League -- League Summary. In Gal A. Kaminka, Pedro U. Lima, and Raul Rojas, editors, RoboCup 2002: Robot Soccer World Cup~VI, volume 2752 of Lecture Notes in Artificial Intelligence, pages 443-452. Springer, 2003.
[Bibtex] [Abstract] [Details]
In the simulation league the RoboCup soccer server provides a standard platform for simulated soccer teams to play against each other over a local network. Each team connects 11 player programs and possibly a coach client to the server, which simulates the 2D soccer field and distributes the sensory information to the clients. Besides the team clients the RoboCup soccer monitor or other visualization and debug tools can be connected as a client to the server to provide 2D or 3D visual information or information like game statistics and analysis for the spectators.

@InCollection{ OP03,
abstract  = {In the simulation league the RoboCup soccer server provides a standard platform for simulated soccer teams to play against each other over a local network. Each team connects 11 player programs and possibly a coach client to the server, which simulates the 2D soccer field and distributes the sensory information to the clients. Besides the team clients the RoboCup soccer monitor or other visualization and debug tools can be connected as a client to the server to provide 2D or 3D visual information or information like game statistics and analysis for the spectators.},
adress  = {Berlin, Heidelberg, New York},
author  = {Oliver Obst and Daniel Polani},
booktitle  = {RoboCup 2002: Robot Soccer World Cup~VI},
editor  = {Gal A. Kaminka and Pedro U. Lima and Raul Rojas},
pages = {443--452},
place = {REGAL},
publisher  = {Springer},
series  = {Lecture Notes in Artificial Intelligence},
title = {Simulation League -- League Summary},
volume  = {2752},
year = {2003}
}

• Heni Ben Amor, Oliver Obst, and Jan Murray. Fast, Neat and Under Control: Inverse Steering Behaviors for Physical Autonomous Agents. Fachberichte Informatik 12--2003, Universität Koblenz-Landau, Universität Koblenz-Landau, Institut für Informatik, Rheinau 1, D-56075 Koblenz, 2003.
[Bibtex] [Abstract] [Details]
Steering behaviors are a set of motion based reactive procedures used for navigating autonomous agents in their environment. Combinations of steering behaviors can be used to create complex behaviors. One problem inherent to existing approaches to arbitrating between single behaviors is that their combination may lead to suboptimal, undesired, or even catastrophic results in certain situations. In our paper we present a solution to these problems by introducing inverse steering behaviors for controlling physical agents. Inverse steering behaviors change the original concept of steering behaviors and facilitate improved arbitration between different options by using cost based heuristics. \par We also show a concrete application of inverse steering behaviors, namely the implementation of a dribbling skill with sophisticated obstacle avoidance for a RoboCup soccer agent.

@TechReport{ AOM03,
abstract  = { Steering behaviors are a set of motion based reactive procedures used for navigating autonomous agents in their environment. Combinations of steering behaviors can be used to create complex behaviors. One problem inherent to existing approaches to arbitrating between single behaviors is that their combination may lead to suboptimal,
undesired, or even catastrophic results in certain situations. In our paper we present a solution to these problems by introducing inverse steering behaviors for controlling physical agents. Inverse steering behaviors change the original concept of steering behaviors and facilitate improved arbitration between different options by using cost based heuristics. \par We also show a concrete application of inverse steering behaviors, namely the implementation of a dribbling skill with sophisticated obstacle avoidance for a RoboCup soccer agent.},
address  = {Universit{\"a}t Koblenz-Landau, Institut f{\"u}r Informatik, Rheinau 1, D-56075 Koblenz},
author  = {Heni {Ben Amor} and Oliver Obst and Jan Murray},
institution  = {Universit{\"a}t Koblenz-Landau},
language  = {english},
number  = {12--2003},
pdf = {http://www.uni-koblenz.de/fb4/publikationen/gelbereihe/RR-12-2003.pdf}
,
title = {Fast, Neat and Under Control: Inverse Steering Behaviors for Physical Autonomous Agents},
type = {Fachberichte Informatik},
year = {2003}
}

• Minoru Asada, Oliver Obst, Daniel Polani, Brett Browning, Andrea Bonarini, Masahiro Fujita, Thomas Christaller, Tomoichi Takahashi, Satoshi Tadokoro, Elizabeth Sklar, and Gal A. Kaminka. An Overview of RoboCup-2002 Fukuoka/Busan. AI Magazine, 24(2):21-40, Summer 2003.
[Bibtex] [Details]
@Article{ AOP+03,
author  = {Minoru Asada and Oliver Obst and Daniel Polani and Brett Browning and Andrea Bonarini and Masahiro Fujita and Thomas Christaller and Tomoichi Takahashi and Satoshi Tadokoro and Elizabeth Sklar and Gal A. Kaminka},
journal  = {AI Magazine},
month = {Summer},
number  = {2},
pages = {21--40},
place = {REGAL},
title = {An Overview of {RoboCup-2002} {Fukuoka}/{Busan}},
volume  = {24},
year = {2003}
}

### 2002

• Oliver Obst. Specifying Rational Agents with Statecharts and Utility Functions. In Andreas Birk, Silvia Coradeschi, and Satoshi Tadokoro, editors, RoboCup-01: Robot Soccer WorldCup~V, number 2377 in Lecture Notes in Artificial Intelligence, pages 173-182. Springer, Berlin, Heidelberg, New York, 2002.
[Bibtex] [Abstract] [Details]
Preliminary version appeared as Fachberichte Informatik 5/2001, Universität Koblenz-Landau.

To aid the development of the robotic soccer simulation league team RoboLog-2000, a method for the specification of multi-agent teams by statecharts has been introduced. The results in the last years competitions showed that though the team was competitive, it did not behave adaptive in unknown situations. The design of adaptive agents with this method is possible, but not in a straightforward manner. The purpose of this paper is to extend the approach by a more adaptive action selection mechanism and to facilitate a more explicit representation of goals of an agent.

@InCollection{ Obst02a,
abstract  = { To aid the development of the robotic soccer simulation league team RoboLog-2000, a method for the specification of multi-agent teams by statecharts has been introduced. The results in the last years competitions showed that though the team was competitive, it did not behave adaptive in unknown situations. The design of adaptive agents with this method is possible, but not in a straightforward manner. The purpose of this paper is to extend the approach by a more adaptive action selection mechanism and to facilitate a more explicit representation of goals of an agent. },
address  = {Berlin, Heidelberg, New York},
author  = {Oliver Obst},
booktitle  = {RoboCup-01: Robot Soccer WorldCup~V},
,
number  = {2377},
pages = {173--182},
pdf = {http://www.uni-koblenz.de/~fruit/PAPERS/obs01a.pdf},
psgz = {http://www.uni-koblenz.de/~fruit/PAPERS/obs01a.ps.gz},
publisher  = {Springer},
series  = {Lecture Notes in Artificial Intelligence},
title = {Specifying Rational Agents with Statecharts and Utility Functions},
wwwnote  = {<a href="http://www.uni-koblenz.de/fb4/publikationen/gelbereihe/RR-5-2001.pdf">Preliminary&nbsp;version</a> appeared as Fachberichte Informatik 5/2001, Universit{\"a}t Koblenz-Landau.},
year = {2002}
}

• Frieder Stolzenburg, Oliver Obst, and Jan Murray. Qualitative Velocity and Ball Interception. In Matthias Jarke, Jana Köhler, and Gerhard Lakemeyer, editors, KI-2002: Advances in Artificial Intelligence -- Proceedings of the 25th Annual German Conference on Artificial Intelligence, number 2479 in Lecture Notes in Artificial Intelligence, pages 283-298, Berlin, Heidelberg, New York, 2002. Springer.
[Bibtex] [Abstract] [Details]
Preliminary version appeared as Fachberichte Informatik 4/2002, Universität Koblenz-Landau.

In many approaches for qualitative spatial reasoning, navigation of an agent in a more or less static environment is considered (e.g. in the double-cross calculus). However, in general, real environment are dynamic, which means that both the agent itself and also other objects and agents in the environment may move. Thus, in order to perform spatial reasoning, not only (qualitative) distance and orientation information is needed, but also information about (relative) velocity of objects. Therefore, we will introduce concepts for qualitative and relative velocity: (quick) to left, neutral, (quick) to right. We investigate the usefulness of this approach in a case study, namely ball interception of simulated soccer agents in the RoboCup. We compare a numerical approach where the interception point is computed exactly, a strategy based on reinforcement learning, a method with qualitative velocities developed in this paper, and the na{\"\i}ve method where the agent simply goes directly to the actual ball position.

@InProceedings{ SOM02b,
abstract  = {In many approaches for qualitative spatial reasoning,
navigation of an agent in a more or less static environment is considered (e.g. in the double-cross calculus). However,
in general, real environment are dynamic, which means that both the agent itself and also other objects and agents in the environment may move. Thus, in order to perform spatial reasoning, not only (qualitative) distance and orientation information is needed, but also information about (relative) velocity of objects. Therefore, we will introduce concepts for qualitative and relative velocity: (quick) to left, neutral, (quick) to right. We investigate the usefulness of this approach in a case study, namely ball interception of simulated soccer agents in the RoboCup. We compare a numerical approach where the interception point is computed exactly, a strategy based on reinforcement learning, a method with qualitative velocities developed in this paper, and the na{\"\i}ve method where the agent simply goes directly to the actual ball position. },
address  = {Berlin, Heidelberg, New York},
author  = {Frieder Stolzenburg and Oliver Obst and Jan Murray},
booktitle  = {KI-2002: Advances in Artificial Intelligence -- Proceedings of the 25th Annual German Conference on Artificial Intelligence},
editor  = {Matthias Jarke and Jana K{\"o}hler and Gerhard Lakemeyer},
,
number  = {2479},
pages = {283--298},
publisher  = {Springer},
series  = {Lecture Notes in Artificial Intelligence},
title = {Qualitative Velocity and Ball Interception},
wwwnote  = {<a href="http://www.uni-koblenz.de/fb4/publikationen/gelbereihe/RR-12-01.pdf">Preliminary&nbsp;version</a> appeared as Fachberichte Informatik 4/2002, Universit{\"a}t Koblenz-Landau.},
year = {2002}
}

• Jan Murray, Oliver Obst, and Frieder Stolzenburg. RoboLog Koblenz 2001. In Andreas Birk, Silvia Coradeschi, and Satoshi Tadokoro, editors, RoboCup 2001: Robot Soccer World Cup~V, volume 2377 of Lecture Notes in Artificial Intelligence, pages 526-530. Springer, Berlin, Heidelberg, New York, 2002. Team description
[Bibtex] [Details]
@InCollection{ MOS02a,
address  = {Berlin, Heidelberg, New York},
author  = {Jan Murray and Oliver Obst and Frieder Stolzenburg},
booktitle  = {RoboCup 2001: Robot Soccer World Cup~V},
,
note = {Team description},
pages = {526--530},
publisher  = {Springer},
series  = {Lecture Notes in Artificial Intelligence},
title = {{R}obo{L}og {K}oblenz 2001},
volume  = {2377},
year = {2002}
}

### 2001

• Jan Murray, Oliver Obst, and Frieder Stolzenburg. Towards a Logical Approach for Soccer Agents Engineering. In Peter Stone, Tucker Balch, and Gerhard Kraetzschmar, editors, RoboCup-2000: Robot Soccer World Cup IV, number 2019 in Lecture Notes in Artificial Intelligence, pages 199-208. Springer, Berlin, Heidelberg, New York, 2001.
[Bibtex] [Abstract] [Details]
Preliminary version appeared as Fachberichte Informatik 6/2000, Universität Koblenz-Landau.

Building agents for a scenario such as the RoboCup simulation league requires not only methodologies for implementing high-level complex behavior, but also the careful and efficient programming of low-level facilities like ball interception. With this hypothesis in mind, we continued the development of RoboLog Koblenz. As before, the focus is laid on the declarativity of the approach. This means, agents are implemented in a logic- and rule-based manner in the high-level and flexible logic programming language Prolog. Logic is used as a control language for deciding how an agent should behave in a situation where there possibly is more than one choice. \par In order to describe the more procedural aspects of the agent's behavior, we employ state machines, which are represented by statecharts. Because of this, we revised our script language for modeling multi-agent behavior, such that we are now able to specify plans with iterative parts and also reactive behavior, which is triggered by external events. In summary, multi-agent behavior can be described in a script language, where procedural aspects are specified by statecharts and declarative aspects by logical rules (in decision trees). Multi-agent scripts are implemented in Prolog. The RoboLog kernel is written in C++ and makes now use of the low-level skills of the CMUnited-99 simulator team.

@InCollection{ MOS01b,
abstract  = {Building agents for a scenario such as the RoboCup simulation league requires not only methodologies for implementing high-level complex behavior, but also the careful and efficient programming of low-level facilities like ball interception. With this hypothesis in mind, we continued the development of RoboLog Koblenz. As before,
the focus is laid on the declarativity of the approach. This means, agents are implemented in a logic- and rule-based manner in the high-level and flexible logic programming language Prolog. Logic is used as a control language for deciding how an agent should behave in a situation where there possibly is more than one choice. \par In order to describe the more procedural aspects of the agent's behavior, we employ state machines, which are represented by statecharts. Because of this, we revised our script language for modeling multi-agent behavior, such that we are now able to specify plans with iterative parts and also reactive behavior, which is triggered by external events. In summary, multi-agent behavior can be described in a script language, where procedural aspects are specified by statecharts and declarative aspects by logical rules (in decision trees). Multi-agent scripts are implemented in Prolog. The RoboLog kernel is written in C++ and makes now use of the low-level skills of the CMUnited-99 simulator team.},
address  = {Berlin, Heidelberg, New York},
author  = {Jan Murray and Oliver Obst and Frieder Stolzenburg},
booktitle  = {{R}obo{C}up-2000: Robot Soccer World Cup {IV}},
editor  = {Peter Stone and Tucker Balch and Gerhard Kraetzschmar},
html = {http://www.coral.cs.cmu.edu/robocup/workshop2000/},
number  = {2019},
pages = {199--208},
place = {REGAL},
psgz = {http://www.uni-koblenz.de/ag-ki/PAPER/ROBOCUP/2000/teamdes.ps.gz}
,
publisher  = {Springer},
series  = {Lecture Notes in Artificial Intelligence},
title = {Towards a Logical Approach for Soccer Agents Engineering},
wwwnote  = {<a href="http://www.uni-koblenz.de/fb4/publikationen/gelbereihe/RR-6-2000.pdf">Preliminary&nbsp;version</a> appeared as Fachberichte Informatik 6/2000, Universit{\"a}t Koblenz-Landau.},
year = {2001}
}

• Jan Murray, Oliver Obst, and Frieder Stolzenburg. RoboLog Koblenz 2000. In Peter Stone, Tucker Balch, and Gerhard Kraetzschmar, editors, RoboCup-2000: Robot Soccer World Cup IV, number 2019 in Lecture Notes in Artificial Intelligence, pages 469-472. Springer, Berlin, Heidelberg, New York, 2001. Team description.
[Bibtex] [Details]
@InCollection{ MOS01a,
address  = {Berlin, Heidelberg, New York},
author  = {Jan Murray and Oliver Obst and Frieder Stolzenburg},
booktitle  = {{R}obo{C}up-2000: Robot Soccer World Cup {IV}},
editor  = {Peter Stone and Tucker Balch and Gerhard Kraetzschmar},
note = {Team description.},
number  = {2019},
pages = {469--472},
place = {REGAL},
psgz = {http://www.uni-koblenz.de/ag-ki/PAPER/ROBOCUP/2000/description.ps.gz}
,
publisher  = {Springer},
series  = {Lecture Notes in Artificial Intelligence},
title = {{R}obo{L}og {K}oblenz 2000},
year = {2001}
}

### 2000

• Frieder Stolzenburg, Oliver Obst, Jan Murray, and Björn Bremer. Spatial Agents Implemented in a Logical Expressible Language. In Manuela Veloso, Enrico Pagello, and Hiroaki Kitano, editors, RoboCup-99: Robot Soccer World Cup~III, volume 1856 of Lecture Notes in Artificial Intelligence. Springer, 2000.
[Bibtex] [Abstract] [Details]
Preliminary version appeared as Fachberichte Informatik 4/1999, Universität Koblenz-Landau.

In this paper, we present a multi-layered architecture for spatial and temporal agents. The focus is laid on the declarativity of the approach, which makes agent scripts expressive and well understandable. They can be realized as (constraint) logic programs. The logical description language is able to express actions or plans for one and more autonomous and cooperating agents for the RoboCup (Simulator League). The system architecture hosts constraint technology for qualitative spatial reasoning, but quantitative data is taken into account, too. The basic (hardware) layer processes the agent's sensor information. An interface transfers this low-level data into a logical representation. It provides facilities to access the preprocessed data and supplies several basic skills. The second layer performs (qualitative) spatial reasoning. On top of this, the third layer enables more complex skills such as passing, offside-detection etc. At last, the fourth layer establishes acting as a team both by emergent and explicit cooperation. Logic and deduction provide a clean means to specify and also to implement teamwork behavior.

@InCollection{ SOMB00,
abstract  = {In this paper, we present a multi-layered architecture for spatial and temporal agents. The focus is laid on the declarativity of the approach, which makes agent scripts expressive and well understandable. They can be realized as (constraint) logic programs. The logical description language is able to express actions or plans for one and more autonomous and cooperating agents for the RoboCup (Simulator League). The system architecture hosts constraint technology for qualitative spatial reasoning,
but quantitative data is taken into account, too. The basic (hardware) layer processes the agent's sensor information. An interface transfers this low-level data into a logical representation. It provides facilities to access the preprocessed data and supplies several basic skills. The second layer performs (qualitative) spatial reasoning. On top of this, the third layer enables more complex skills such as passing, offside-detection etc. At last, the fourth layer establishes acting as a team both by emergent and explicit cooperation. Logic and deduction provide a clean means to specify and also to implement teamwork behavior.},
adress  = {Berlin, Heidelberg, New York},
author  = {Frieder Stolzenburg and Oliver Obst and Jan Murray and Bj{\"o}rn Bremer},
booktitle  = {RoboCup-99: Robot Soccer World Cup~III},
editor  = {Manuela Veloso and Enrico Pagello and Hiroaki Kitano},
place = {REGAL},
publisher  = {Springer},
series  = {Lecture Notes in Artificial Intelligence},
title = {Spatial Agents Implemented in a Logical Expressible Language},
volume  = {1856},
wwwnote  = {<a href="http://www.uni-koblenz.de/fb4/publikationen/gelbereihe/RR-4-99.pdf">Preliminary&nbsp;version</a> appeared as Fachberichte Informatik 4/1999, Universit{\"a}t Koblenz-Landau.},
year = {2000}
}

• Ulrich Furbach, Oliver Obst, and Frieder Stolzenburg. Intelligente Agenten und KI. LOG~IN -- Informatische Bildung und Computer in der Schule, 20(3/4):17-21, 2000.
[Bibtex] [Abstract] [Details]
Neuere KI-Textbücher stützen sich zur Begriffsdefinition von "Künstliche Intelligenz" in der Regel massiv auf den Agentenbegriff. Autonomie, Körperhaftigkeit (embodiment), Reaktivität und Situiertheit in einem komplexen Kontext sind unmittelbar mit den Begriffen Agenten und KI verwoben. Hat man in der Vergangenheit eher versucht, einzelne Maschinen mit mächtigen Wissensverarbeitungsmechanismen auszustatten, wird aus heutiger Sicht auf die Interaktion mit der Umwelt und anderen Agenten gezielt. Natürlich muss auch hierbei Wissen repräsentiert und verarbeitet werden. Wir skizzieren die gängigsten Architekturprinzipien für Agenten und gehen unter anderem auf logikbasierte und BDI-Architekturen (BDI = Belief-Desire-Intention) näher ein. Außerdem werden zwei prinzipiell verschiedene Anwendungsmöglichkeiten an Beispielen beschrieben: Zum einen können Agenten benutzt werden, um Wissen zu beschaffen und aufzubereiten; die Verfügbarkeit von Wissen stellt zum anderen eine Voraussetzung für intelligentes Verhalten von autonomen Robotern dar. Dies wird veranschaulicht durch fußballspielende autonome Agenten im RoboCup und durch die Betrachtung eines Systems zur Informationsextraktion im Internet. ß

@Article{ FOS00,
abstract  = {Neuere KI-Textb{\"u}cher st{\"u}tzen sich zur Begriffsdefinition von "K{\"u}nstliche Intelligenz" in der Regel massiv auf den Agentenbegriff. Autonomie,
K{\"o}rperhaftigkeit (embodiment), Reaktivit{\"a}t und Situiertheit in einem komplexen Kontext sind unmittelbar mit den Begriffen Agenten und KI verwoben. Hat man in der Vergangenheit eher versucht, einzelne Maschinen mit m{\"a}chtigen Wissensverarbeitungsmechanismen auszustatten,
wird aus heutiger Sicht auf die Interaktion mit der Umwelt und anderen Agenten gezielt. Nat{\"u}rlich muss auch hierbei Wissen repr{\"a}sentiert und verarbeitet werden. Wir skizzieren die g{\"a}ngigsten Architekturprinzipien f{\"u}r Agenten und gehen unter anderem auf logikbasierte und BDI-Architekturen (BDI = Belief-Desire-Intention) n{\"a}her ein. Au{\ss}erdem werden zwei prinzipiell verschiedene Anwendungsm{\"o}glichkeiten an Beispielen beschrieben: Zum einen k{\"o}nnen Agenten benutzt werden,
um Wissen zu beschaffen und aufzubereiten; die Verf{\"u}gbarkeit von Wissen stellt zum anderen eine Voraussetzung f{\"u}r intelligentes Verhalten von autonomen Robotern dar. Dies wird veranschaulicht durch fu{\ss}ballspielende autonome Agenten im RoboCup und durch die Betrachtung eines Systems zur Informationsextraktion im Internet. {\ss}},
author  = {Ulrich Furbach and Oliver Obst and Frieder Stolzenburg},
journal  = {LOG~IN -- Informatische Bildung und Computer in der Schule},
number  = {3/4},
pages = {17-21},
place = {REGAL},
title = {Intelligente {A}genten und {KI}},
volume  = {20},
year = {2000}
}

• Jan Murray, Oliver Obst, and Frieder Stolzenburg. RoboLog Koblenz 2000. In Wiebe van der Hoek, editor, Proceedings of the Workshop for the Robocup European Championship. Vrije Universiteit Amsterdam,, 2000. Team description
[Bibtex] [Details]
@InProceedings{ MOS00a,
author  = {Jan Murray and Oliver Obst and Frieder Stolzenburg},
booktitle  = {Proceedings of the {Workshop for the Robocup European Championship}},
editor  = {Wiebe van der Hoek},
html = {http://www.cs.uu.nl/people/wiebe/Robocup/},
note = {Team description},
organization  = {Vrije Universiteit Amsterdam},
title = {{R}obo{L}og {K}oblenz 2000},
year = {2000}
}

• Jan Murray, Oliver Obst, and Frieder Stolzenburg. RoboLog Koblenz. In Manuela Veloso, Enrico Pagello, and Hiroaki Kitano, editors, RoboCup-99: Robot Soccer World Cup~III, volume 1856 of Lecture Notes in Artificial Intelligence, pages 481-494. Springer, 2000. Team Description
[Bibtex] [Details]
@InCollection{ MOS00b,
adress  = {Berlin, Heidelberg, New York},
author  = {Jan Murray and Oliver Obst and Frieder Stolzenburg},
booktitle  = {RoboCup-99: Robot Soccer World Cup~III},
editor  = {Manuela Veloso and Enrico Pagello and Hiroaki Kitano},
note = {Team Description},
pages = {481--494},
place = {REGAL},
publisher  = {Springer},
series  = {Lecture Notes in Artificial Intelligence},
title = {{R}obo{L}og {K}oblenz},
volume  = {1856},
year = {2000}
}

### 1999

• Jan Murray, Frieder Stolzenburg, Oliver Obst, and Björn Bremer. RoboLog Koblenz: Complex Agent Scripts Implemented in Logic. In Stefan Sablatnög and Stefan Enderle, editors, Proceedings of the Workshop RoboCup during KI'99 in Bonn, pages 12-25, 1999. SFB~527 Report 1999/12, Universität Ulm
[Bibtex] [Details]
@InProceedings{ MSO+99,
author  = {Jan Murray and Frieder Stolzenburg and Oliver Obst and Bj{\"o}rn Bremer},
booktitle  = {Proceedings of the {Workshop RoboCup} during {KI'99} in Bonn},
editor  = {Stefan Sablatn{\"o}g and Stefan Enderle},
note = {SFB~527 Report 1999/12, Universit{\"a}t Ulm},
pages = {12--25},
title = {{RoboLog Koblenz}: Complex Agent Scripts Implemented in Logic},
year = {1999}
}

• Oliver Obst and Frieder Stolzenburg. Der RoboCup während der IJCAI'99. KI, 13(4):66-67, 1999. Tagungsbericht
[Bibtex] [Details]
@Article{ OS99,
author  = {Oliver Obst and Frieder Stolzenburg},
journal  = {KI},
note = {Tagungsbericht},
number  = {4},
pages = {66--67},
title = {Der {R}obo{C}up w{\"a}hrend der {IJCAI}'99},
volume  = {13},
year = {1999}
}

• Oliver Obst. RoboLog: Eine deduktive Schnittstelle zum RoboCup Soccer Server. Master's thesis, Universität Koblenz, feb 1999.
[Bibtex] [Details]
@MastersThesis{ Obst99b,
author  = {Oliver Obst},
month = feb,
place = {REGAL},
psgz = {http://www.uni-koblenz.de/~fruit/PAPERS/thesis.ps.gz},
school  = {Universit{\"a}t Koblenz},
title = {{R}obo{L}og: {E}ine deduktive {S}chnittstelle zum {R}obo{C}up {S}occer {S}erver},
year = {1999}
}

• Frieder Stolzenburg, Oliver Obst, Jan Murray, and Björn Bremer. RoboLog Koblenz: Spatial Agents Implemented in a Logical Expressible Language. In Silvia Coradeschi, Tucker Balch, Gerhard Kraetzschmar, and Peter Stone, editors, Team Descriptions --- Simulation League, pages 116-120. Linköping University Electronic Press, 1999.
[Bibtex] [Abstract] [Details]
In this paper, we present a multi-layered architecture for spatial and temporal agents. The focus is laid on the declarativity of the approach, which makes agent scripts expressive and well understandable. They can be realized as (constraint) logic programs. The logical description language is able to express actions or plans for one and more autonomous and cooperating agents for the RoboCup (Simulator League). The system architecture hosts constraint technology for qualitative spatial reasoning, but quantitative data is taken into account, too. The basic (hardware) layer processes the agent's sensor information. An interface transfers this low-level data into a logical representation. It provides facilities to access the preprocessed data and supplies several basic skills. The second layer performs (qualitative) spatial reasoning. On top of this, the third layer enables more complex skills such as passing, offside-detection etc. At last, the fourth layer establishes acting as a team both by emergent and explicit cooperation. Logic and deduction provide a clean means to specify and also to implement teamwork behavior.

@InProceedings{ SOMB99b,
abstract  = { In this paper, we present a multi-layered architecture for spatial and temporal agents. The focus is laid on the declarativity of the approach, which makes agent scripts expressive and well understandable. They can be realized as (constraint) logic programs. The logical description language is able to express actions or plans for one and more autonomous and cooperating agents for the RoboCup (Simulator League). The system architecture hosts constraint technology for qualitative spatial reasoning,
but quantitative data is taken into account, too. The basic (hardware) layer processes the agent's sensor information. An interface transfers this low-level data into a logical representation. It provides facilities to access the preprocessed data and supplies several basic skills. The second layer performs (qualitative) spatial reasoning. On top of this, the third layer enables more complex skills such as passing, offside-detection etc. At last, the fourth layer establishes acting as a team both by emergent and explicit cooperation. Logic and deduction provide a clean means to specify and also to implement teamwork behavior. },
author  = {Frieder Stolzenburg and Oliver Obst and Jan Murray and Bj{\"o}rn Bremer},
booktitle  = {Team Descriptions --- Simulation League},
editor  = {Silvia Coradeschi and Tucker Balch and Gerhard Kraetzschmar and Peter Stone},
html = {http://www.ida.liu.se/ext/robocup/simul/RobologKoblenz99/teampage.html}
,
pages = {116--120},
publisher  = {Link{\"o}ping University Electronic Press},
title = {{RoboLog Koblenz}: Spatial Agents Implemented in a Logical Expressible Language},
year = {1999}
}

### 1998

• Oliver Obst, Jan Murray, Frieder Stolzenburg, and Björn Bremer. Towards Deduction in RoboCup. In Proceedings of the RoboCup Workshop during KI'98, Bremen, 1998.
[Bibtex] [Details]
@InProceedings{ OMS+98,
author  = {Oliver Obst and Jan Murray and Frieder Stolzenburg and Bj{\"o}rn Bremer},
booktitle  = {Proceedings of the {RoboCup Workshop} during {KI'98}},
html = {http://www.uni-koblenz.de/~fruit/PAPERS/ROBOCUP/TOWARDS/index.html}
,
title = {Towards Deduction in {R}obo{C}up},
url = {http://www.ki.informatik.hu-berlin.de/AKRoboCup/robocup-ki98.html}
,
year = {1998}
}

• Oliver Obst. RoboCup: FC-Linux - Mit Linux zur Fußball-WM. Linux-Magazin, (8):48-51, 1998.
[Bibtex] [Details]
@Article{ Obst98a,
author  = {Oliver Obst},
html = {http://www.linux-magazin.de/Artikel/ausgabe/1998/08/RoboCup/robocup.html}
,
journal  = {Linux-Magazin},
number  = {8},
pages = {48--51},
place = {Regal},
title = {{R}obo{C}up: {FC}-{L}inux - {M}it {L}inux zur {F}u{\ss}ball-{WM}},
year = {1998}
}

• Oliver Obst. RoboLog -- An ECLiPSe-Prolog SoccerServer Interface: Users Manual. , mar 1998.
[Bibtex] [Details]
@Manual{ Obst98b,
author  = {Oliver Obst},
month = mar,
title = {{R}obo{L}og -- {A}n {ECLiPSe}-{P}rolog {S}occer{S}erver Interface: Users Manual},
year = {1998}
}

### 1997

• Frieder Stolzenburg and Oliver Obst. Reasoning with Constraints and Well-Founded Negation. In Documents of the ERCIM/Compulog Workshop on Constraints, Schloß Hagenberg, Linz, 1997.
[Bibtex] [Details]
An abstract appeared in Fachberichte Informatik 23/1997, pp. 9-10, Universität Koblenz-Landau
@InProceedings{ SO97b,
}