Motionplanning For Car-like Robots Using Probabilistic Learning Approach With Heuristic Nodes Enhancement

The motion planning problem asks for determine a collision-free feasible path for arobot amidst a set of obstacles. As traditional Probabilistic Roadmap Methods are notfeasible enough and seem to non-reusable, we present a new approach for solving thisproblem, based on traditional PRMs with heuristic nodes enhancement, using cubicB-splines method in the optimization and smoothing of the paths found after queryphase. Compared to traditional PRMs, major advantages of our approach are listed asfollows. First, in the construction stage of roadmap, we build roadmap with efficientnodes enhancement which reduces unless computation and processes narrow passage.Second, we do not need the exact car-like robot’s turning radii before query phrase,which means our approach is very feasible to different turning radii.Our approach applies different techniques in the optimization of different stages ofPRMs to gain overall efficiency enhancement. Our approach makes changes in twoways. First, we use nodes enhancement strategies during the construction of roadmap.Second, we try to optimize the shortest path found after query phase with cubicB-splines methods.In this work, all experiments are performed with C++language. We show someresults obtained using our PRM approach for non-holonomic car-like robots indifferent scenes. Results with our approach are very plausible. Analysis ofexperimental data led to better strategy and appropriate value of parameters.