Research On Improved Robot Path Planning Algorithm Based On Rapidly-exploring Random Tree

With the wide application of robot in many fields,some common repetitive and dangerous work can be done by robot.As an important research content of robot,path planning has been paid much attention by domestic and foreign scholars.As a novel random node sampling algorithm,the Rapidly-exploring Random Tree algorithm(RRT)has been greatly researched and developed since it was proposed.Compared with the traditional path planning algorithm,RRT has the advantages of short modeling time,strong search ability and convenient addition of nonholonomic constraints,but the random expansion of nodes results in poor path quality and cannot converge to the global optimum.RRT*,the variant of RRT,can generate the optimal path,but it takes a long time.Based on RRT,this paper proposes the SRRT algorithm based on reinforcement learning.In the process of node expansion,Markov decision modeling is adopted,and Sarsa algorithm is used to train nodes,so as to improve the path quality.And for its variant RRT*,an AIRRT* algorithm based on homotopy asymptotic optimization is proposed,which combines ellipse region sampling and homotopy path optimization process to improve the convergence speed of the algorithm.The main research contents and innovation points of this paper are as follows:(1)The SRRT algorithm based on reinforcement learning.For RRT node random extend,using Sarsa round status update mechanism instead of single step,establish a Markov model,define the algorithm of the state and action Spaces,the variable parameter greedy strategy and continuous feedback reward function training node extension process,through the Q meter to choose effective action strategy,and generate a suboptimal convergence path with stable path value.Simulation results show that the path value after SRRT convergence is shorter than the initial average path value in different environments.The path quality and success searching rate after SRRT convergence is better than RRT in a certain number of iterations.These results reflect the convergence effect and overall optimization performance of SRRT.(2)The AIRRT* algorithm based on homotopy asymptotic optimization,which combines the elliptic sampling area with the homotopy path optimization process,and introduces the elliptic region sampling principle of IRRT* algorithm in detail.On the basis of IRRT*,adds the homotopy path asymptotic optimization process.Through linear interpolation and adaptive selection of environment information points,the original path is optimized as homotopy optimal path,the ellipse sampling area is reduced,and the convergence speed of the optimal path is improved.Simulation results show that AIRRT* converges to the optimal path faster than IRRT* and RRT* in different environments,which shows its ability to quickly converge to the optimal path.