RoboCup Simulation Leagues: Enabling Replicable and Robust Investigation of Complex Robotic Systems

Physically realistic simulated environments are powerful platforms for enabling measurable, replicable, and statistically robust investigation of complex robotic systems. Such environments are epitomized by the RoboCup (RC) simulation leagues, which have been successfully utilized to conduct massively parallel experiments on a variety of topics, including optimization of bipedal locomotion, self-localization from noisy perception data, and planning complex multiagent strategies without direct agent-to-agent communication. Many of these systems are later transferred to physical robots, making the simulation leagues invaluable beyond the scope of simulated soccer matches.