Research On Monte Carlo Self-Localization Algorithm For Soccer Robot In Robocup Humanoid League

As the robot applications domain expands ceaselessly, people hope that the robot canwork at a wider range environment and the behavior of robot is more intelligent, whichrequires the robot in the environment has a stronger adaptability and a stronger ability tointeract with the environment. The competition environment of humanoid robot soccer isdynamic, uncertain and multi-agent, therefore, whether the design of software structure orthe control algorithms for robot system are both proposed higher requirements. Thereal-time, accurate positioning is the key to improve the performance of the robot, and it’salso a precondition for robot to complete path planning, navigation, decision-making andtask execution. Taking soccer robot in RoboCup humanoid league as the researchbackground, this paper focuses on the study of Monte Carlo localization method which isbased on probability, the main research work is as follows:(1) This paper established related models for the control system of robot and a map ofthe environment, the models are based on the humanoid robot platform Erectus, thenaccording to the rules of RoboCup soccer humanoid league to design the task environmentof soccer robot system, which laid the model foundation for the research of the probabilisticlocalization algorithms.(2) For the traditional Monte Carlo localization algorithm has some problems, such asparticles running out, an improved Monte Carlo localization algorithm is proposed in thispaper, adopting new resampling rules to ensure the state space always has a few particlesdistributed randomly in the process of robot localization, increase the diversity of particles.The experimental results show that the algorithm is efficient, meanwhile, it can solve boththe particle running out and the kidnapped robot problem, improving the performance ofthe traditional Monte Carlo localization algorithm.(3) For the high real-time requirement of soccer robot system, this paper use thesubregion Monte Carlo localization algorithm to control the particle filter parameters andresampling rules in different domains, thus the real-time performance is improved. To solvethe problem of particles resource waste in a localization algorithm with fixed sample size, aself-adaptive subregion Monte Carlo localization algorithm is proposed in this paper, the KL distance is applied to the subregion Monte Carlo localization algorithm, using KLDsampling method and adjustting the number of sampling particles adaptively, determiningthe effective sample size. The experimental results show that the self-adaptive subregionMonte Carlo localization algorithm has the advantages of high accuracy of positioning andfast running efficiency, avoiding the waste of particles resource.(4) To solve the kidnapped robot problem that may occur in the process of the robotpositioning, a kind of states replacement method is proposed in this paper, researching arecognition strategy to reveal the kidnapped robot problem inmmediately, then using acontrol strategy to realize the robot relocalization. The experimental results show that thestates replacement method can effectively solve the kidnapped robot problem, realizingrapidly the robot relocalization.In summary, the improved self-adaptive Monte Carlo localization algorithm thatproposed in this paper can effectively solve the particle running out and the kidnapped robotproblem, improving the performance of the localization algorithm, so that the robot canbetter complete the task in the competition.