Research On Complete Coverage Path Planning Of Lawn Mower Robot Based On Reinforcement Learning

With the development of intelligent robot technology and the greening industry,the lawn mowing robot industry has made great progress.The Coverage Path Planning(CPP)algorithm,which seeks a path that covers all target points,is an important research direction for autonomous lawn mowing robots.Currently,most lawn mowing robots on the market suffer from low coverage efficiency,low coverage rate,high repetition rate,and high turning frequency,while low-precision positioning systems cause deviations in the coverage path.To address these issues,this paper proposes a lawn mowing robot platform based on fused positioning and an improved Q-Learning algorithm for coverage path planning.The positioning accuracy of the lawn mowing robot is crucial for coverage path planning.To achieve high-precision positioning,this paper proposes a fused positioning technology based on Extended Kalman Filter(EKF)filtering of Ultra Wide Band(UWB)and Inertial Navigation System(INS)positioning.In the fused positioning algorithm,the deviation between UWB and INS positioning information is used as a state value to establish a prediction equation,and the squared difference between the UWB ranging value and the distance value of the INS positioning to the UWB base station is used as an observation value.A dynamic dimension observation matrix is established based on the non-line-of-sight error of the UWB base station,and the EKF filter is used to obtain the optimal estimation value to correct the UWB positioning and improve the accuracy of the positioning system.Simulation results show that the average error of the EKF-based UWB/INS fused positioning is reduced by 30% compared to using UWB alone.To solve the coverage path planning problem of the lawn mowing robot,a coverage map is constructed using grid methods,and an improved Q-Learning algorithm for coverage path planning is proposed by introducing rewards for moving away from the starting point,walking in a straight line,and covering behavior,while considering the influence of coverage rate,repetition rate,and turning frequency on coverage tasks.Simulation results show that the improved Q-Learning algorithm for coverage path planning has lower repetition rate and turning frequency than traditional boustrophedon cellular decomposition algorithms and Q-Learning-based coverage path planning algorithms.A software and hardware experimental platform is built to conduct experimental tests in a real grass environment to validate the proposed EKF-based UWB/INS fused positioning algorithm and improved Q-Learning algorithm for coverage path planning.