Partition Sampling Path Planning Strategy Based On POMDP

With the rapid development of robot science,mobile robots gradually enter human life and play a more and more important role.As a key technology of mobile robot motion,path planning has been paid close attention by experts and scholars in related fields in recent years,and various mature algorithms such as RRT and PRM have been developed.However,most of the algorithms do not consider the uncertainty encountered during the movement of the mobile robot,which leads to the collision of the mobile robot just because of the slight wheel slip.Therefore,only by fully considering the uncertainty,the path planning algorithm can be more reliable applied to the real scene.To this end,this thesis has carried out in-depth research on path planning technology and path planning under uncertain environment.Firstly,a comparative study of several traditional path planning algorithms was carried out,and their respective advantages and disadvantages were analyzed.Finally,the RRT algorithm was selected as the basic path planning algorithm of this thesis.Aiming at the low sampling efficiency of RRT algorithm,a partitioned sampling strategy is proposed.Based on importance sampling,the planning space is divided into open area and narrow area,and the sampling strategy is dynamically adjusted according to the area.When the robot runs in the open area,the open area is uniformly sampled.When the robot runs in the narrow region,Gaussian sampling is performed in the narrow region.Through the reasonable matching of sampling strategy and sampling area,the sampling efficiency of the RRT algorithm is effectively improved,which lays a good foundation for the follow-up work.Aiming at the problem that the RRT algorithm does not consider the uncertainty,a path planning method under uncertainty based on Partially Observable Markov Decision Process(POMDP)was proposed.The method follows the general POMDP solution framework,takes the initial trajectory obtained by the RRT algorithm as input,and uses the iterative linear quadratic Gaussian method to perform value iteration on the belief space,so as to obtain the local optimal solution around the initial trajectory in the belief space,which can also be called the local optimal control strategy,and the new trajectory can be obtained by executing the control strategy,and the appeal process is iterated.Until the final trajectory is obtained.Through simulation,it can be confirmed that the method proposed in this thesis can effectively solve the uncertainty problem in the process of path planning,and has important significance for the practical application of path planning technology.