Software System Design And Trajectory Learning For Small Size Soccer Robots

Small Size Soccer Robot is an important league in Robot World Cup, and it’s also an excellent platform for robot researching, such as the robot system design, multi-robot cooperation, motion planning and trajectory generation. This thesis is based on Small Size League Robot in Zhejiang University, discusses the issues about the multi-robot cooperation in software systems and trajectory learning. Detailed works and results are as following:(1) A software architecture for centralized-controlled multi-robot cooperation has been designed and implemented. This system consists of a Bayes-based evaluation mechanism, the cooperation level, the task level and the execution level. This system solved the long-existed problem, such as module coupling.(2) A framework for extending tactics based on Lua scripts has been designed and implemented. The Lua scripts make it easy to extend the cooperation strategy and robot task. The framework also takes into account a role matching module and a general finite-state-machine module for cooperation level, and a behavior tree based mechanism for task level. This framework makes it easy for switching strategies according to different opponents.(3) A style-adaptive trajectory learning method has been designed and implemented. It uses the demonstrated trajectories as input, and consists of a point distribution model, principal trajectory generation and model parameters training. The method generated by this method is adaptive and robust for goal changing.The system and methods mentioned above been successfully applied into ZJUNlict team, and help to get the champion in RoboCup 2013 and RoboCup 2014.