Omni-directional Walking Skill And Cooperation Mechanism Of Biped Robots In RoboCup3D Simulation Environment

This paper mainly describes the soccer robot individual action technique design and multi-agent cooperation policy assign of humanoid robot soccer team called Apollo3 D. In this paper, it individually realizes omni-directional walking, cooperation localization and team coordination mechanism of biped robot in RoboCup3 D simulation environment. The three main key components as shown below:Firstly, the paper mainly introduces the omni-directional walking motion design of humanoid robot soccer team called Apollo3 D. This section employs a model which is based on double linear inverted pendulum with a predictive control to generate a motion trajectory of the robot’s trunk in the premise of keeping dynamic balance of robots. Parameters for walk are optimized for maximum speed through an approach of Covariance Matrix Adaptation Evolution Strategy(CMA-ES). Facing the difficulties of over fitting in single training task when optimizing the walking skill, this section designs a layered learning strategy using multiple subtasks to enhance walking, turning and dribbling behaviors of the biped robot. With this walking skill, ultimately, rapidly and stably omni-directional walk of biped robot are realized in complex and dynamic environment.Secondly, under the background of gyroscope localization, the localization of three marks and particle filter in RoboCup3 D simulation of the single robot localization, this section employs a new localization method based on improved particle swarm optimization algorithm(PSO). Though combined PSO algorithm with particle filter used in team Apollo3 D localization, the speed and position information of biped robot is updated. At the same time, this section also uses the relative observations of robot to adjust the proposal distribution and the weight of the particles. In the special model of kidnapping for robot, two parameters sloww and fastw are used to track the particle weight average long-term and short-term changes. With this method above, ultimately, the cooperation localization of football robot is realized.Fianlly, in design of the team coordination mechanism, this section proposes a method based on reinforcement learning to realize multi-agent cooperation strategy under the platform of Keepaway. The platform of Keepaway3 vs. 2 was a local tactical strategy in 2D simulation competition proposed by Peter Stone. In this section, it builds a Keepaway model in Robo Cup3 D simulation game, then uses Sarsa( ?) aglothm based linear function approximation in this training model to implement the strategy of soccer robot in team collaboration mechanism, include PASS and GETOPEN strategy.