Dataset archive

RoboCupSimData

Download game state, player perceptions and original simulator logs from 1,125 RoboCup soccer simulation games.

← Back to the RoboCupSimData project page

Three sizes, two formats

This page is part of the RoboCupSimData project.

There are three sets of different sizes on this page:

  1. a one-game set, to explore the data before downloading the full set;
  2. a ten-game set, for more exploring and simple experiments; and
  3. the full dataset, split into ten parts. The full dataset also contains all files included in the one-game and ten-game sets.

All game datasets on this page come in two versions:

  1. A set of CSV files. For each game, there is one file with ground-truth data for all players, one file with game parameters, and forty-four files (two per player) containing each player's local, incomplete and noisy visual perceptions: one file for sensed moving objects and one for sensed landmarks. We modified the soccer simulator to record these local perceptions for every player on both teams; the original soccer simulator does not record them.
  2. The original RoboCup Soccer Simulator logfiles, together with the players' sensory-input logs generated with our simulator patch, including visual and body sensors. The released scripts can recreate all CSV files from these logs.

The file landmarks.csv contains the positions of all landmarks—flags, lines and goals—in global coordinates. Every game uses this set of landmarks, and their global positions are known to all players.

One-game example

An example game between Gliders2016 and HELIOS2016, winner and runner-up of RoboCup 2016.

Ten games from the top five teams

For this ten games top five set, we determined the top five teams from all games run for the full dataset. We then selected ten games so that each team plays each of the other top-five teams once.

Ten teams, 25 rounds

The full dataset was created by playing 25 rounds with ten teams: the top six teams from RoboCup 2017 and the top four teams from RoboCup 2016. In each round, all teams play against each other, resulting in 25 × 9 × 10 ÷ 2 = 1,125 games.

The dataset is split into ten packages for easier downloading. Each package contains every match of one team playing on the left side of the field: four or five match-ups, each played over 25 rounds.

CSV files

17GB zipped / 229GB unzipped
  1. 0-cyrus2017-csv.zip (1.8GB zipped, 24GB unzipped). Content listing: 0-cyrus2017-csv-ls-gGR
  2. 1-fra-united2017-csv.zip (1.6GB zipped, 24GB unzipped). Content listing: 1-fra-united2017-csv-ls-gGR
  3. 2-Gliders2016-csv.zip (1.5GB zipped, 21GB unzipped). Content listing: 2-Gliders2016-csv-ls-gGR
  4. 3-helios2017-csv.zip (1.9GB zipped, 26GB unzipped). Content listing: 3-helios2017-csv-ls-gGR
  5. 4-helios2016-csv.zip (1.5GB zipped, 21GB unzipped). Content listing: 4-helios2016-csv-ls-gGR
  6. 5-hfutengine2017-csv.zip (1.8GB zipped, 23GB unzipped). Content listing: 5-hfutengine2017-csv-ls-gGR
  7. 6-mt2017-csv.zip (1.8GB zipped, 22GB unzipped). Content listing: 6-mt2017-csv-ls-gGR
  8. 7-oxsy2017-csv.zip (2.0GB zipped, 29GB unzipped). Content listing: 7-oxsy2017-csv-ls-gGR
  9. 8-ri-one-csv.zip (1.6GB zipped, 22GB unzipped). Content listing: 8-ri-one-csv-ls-gGR
  10. 9-yunlu-csv.zip (1.5GB zipped, 21GB unzipped). Content listing: 9-yunlu-csv-ls-gGR

Original logs

Simulator and sensory-input logs
  1. 0-cyrus2017-rcx.tar.gz (2.7GB). Content listing: 0-cyrus2017-rcx-ls-gGR
  2. 1-fra-united2017-rcx.tar.gz (2.4GB). Content listing: 1-fra-united2017-rcx-ls-gGR
  3. 2-Gliders2016-rcx.tar.gz (2.1GB). Content listing: 2-Gliders2016-rcx-ls-gGR
  4. 3-helios2017-rcx.tar.gz (2.7GB). Content listing: 3-helios2017-rcx-ls-gGR
  5. 4-helios2016-rcx.tar.gz (2.1GB). Content listing: 4-helios2016-rcx-ls-gGR
  6. 5-hfutengine2017-rcx.tar.gz (2.6GB). Content listing: 5-hfutengine2017-rcx-ls-gGR
  7. 6-mt2017-rcx.tar.gz (2.6GB). Content listing: 6-mt2017-rcx-ls-gGR
  8. 7-oxsy2017-rcx.tar.gz (2.9GB). Content listing: 7-oxsy2017-rcx-ls-gGR
  9. 8-ri-one-rcx.tar.gz (2.3GB). Content listing: 8-ri-one-rcx-ls-gGR
  10. 9-yunlu-rcx.tar.gz (2.3GB). Content listing: 9-yunlu-rcx-ls-gGR