Design And Implementation Of RoboCup Humanoid Soccer Robot Debugging Platform

Humanoid Robot has similar body structure and perception system to humans,whose research involves many academic disciplines such as mechanical engineering,control engineering,electronic information engineering and computer engineering,which is an important research direction in the field of intelligent robots.This paper analyzes the requirements of the upper-level debugging system according to the Robo Cup 2020 Kid Size competition rules and the existing software and hardware framework.Until then,the environmental awareness system took too long to collect environmental data,the action generation system had too many algorithm parameters and the algorithm of the policy generation system is not ideal.So,in this paper,the related algorithms are redesigned to address these main problems 。This paper proposes that the upper-level debugging system consists of three modules: video annotation,gait planning,and strategy generation,which can help participants better obtain the prospective performance of the robot.Video Annotation Module: The application of convolutional neural network greatly improves the accuracy of the robot’s identification of surrounding information,but the training of network models often requires a large amount of sample data.Therefore,this paper uses the image tracking algorithm to let the robot track the target when collecting data,and generate the data set required for the subsequent robot competition at the same time.Gait Planning Module:The robot’s original gait planning algorithm is to generate the motion curve through the gait description parameters firstly,and then generate the gait action by calculating the motion joints of the robot.The advantage of this method is that it is simple to implement,but due to its too many gait parameters,the generated gait movements are difficult to debug,and it takes a lot of time to correct parameters during competition.In this paper,the ZMP theory and the linear inverted pendulum model are used to redesign the gait generation algorithm,and a pressure acquisition board is designed to feedback the state of the robot’s zero moment point to form a closed-loop control of the robot’s gait trajectory.Policy Generation Module: The original decision generation system has the problem that the generation method of the finite-state machine it uses is difficult to fully describe the robot’s game scene,and the boundary parameters of the finite-state machine are difficult to calibrate.This paper models the robot competition scene based on the Markov decision process and builds the Webots training environment.Based on the reinforcement learning algorithm,the value function is generated for the robot strategy,and the strategy generation results are verified on the simulation platform.In conclusion,this paper tests the functions of each module of the upper-level debugging system through several experiments.In the video annotation module,through the test of the actual data of the game scene,the annotation accuracy rate of 85.7% was obtained in the case of using only 1/4 of the time of the original method.Therefore,the method greatly saves data collection time under the condition of meeting the requirements of the competition,so the method is proved effective.In the gait planning module,by testing the real robot,the robot can achieve a walking speed of about 0.165m/s,so the algorithm is verified to be feasible.In the strategy generation module,through the simulation platform experiment,the experimental results show that the strategy generation module can adjust the kicking strategy according to the position dynamics of the opponent’s robot,and the robot completes the organization attack more efficiently than the original strategy generation algorithm.And in the 30 min football match test,it obtained a good score of 8 points ahead of the original strategy generation algorithm.