WrightEagle 2D Soccer Simulation TeamMulti-Agent Systems Lab., USTC| Home | Contests | Members | Publications | Code | Training | Links | Contact | |
WrightEagle 2D Soccer Simulation Team is a branch of WrightEagle Robocup Team, established in 1998 by the Multi-Agent Systems Lab., University of Science and Technology of China (USTC). We have participated in annual competitions of RoboCup since 1999 and have won 6 world champions and 5 runners-up of RoboCup since 2005.
We model RoboCup 2D soccer simulation as a typical problem of partially observable stochastic games (POSGs). In order to solve it online under realtime constraints, we take advantage of rationality assumption, hierarchical decomposition, state abstraction, expectimax tree search, Monte Carlo simulation and heuristic evaluation. More precisely, we model the RoboCup 2D game as a Markov decision process (MDP), assuming that opponents and teammates are all rational. We further decompose the sequential decision-making problem within the resulting MDP into a hierarchy of subtasks, including attack, defense, shoot, dribble, pass, intercept, position, mark, block, formation, etc. A hierarchical online planning algorithm is then developed for the agent to do online planning in realtime, exploiting subtask-specific heuristic and evaluation functions. More information regarding our approach to RoboCup 2D can be found in our recent publications.
A video from RoboCup 2015 final – HELIOS2015 v.s. WrightEagle2015 – is as follows.