A Nonuniform Sampling Strategy for Path Planning Using Heuristic-based Certificate Set

Collision checking is considered one of the most time-consuming tasks in sampling-based path planning algorithms. Traditionally, sampling-based path planning methods draw samples uniformly from the state space, which samples a lot of useless states regardless of previous successful experience. Besides, the sampled states do not contribute to subsequent sampling. However, using the information recorded by sampled states can avoid numerous unnecessary collision checks. Extra information such as previous successful experiences and the distance between the random sample and obstacles should be considered to accelerate the planning process. Thus, it is desirable to devise a nonuniform sampling strategy to utilize the extra information in the planning process. This paper proposes a novel nonuniform sampling strategy maintaining a heuristic-based certificate set during the planning process. The heuristic-based certificate set consists of sampled states with collision status and the minimum distance to the nearest obstacle, while the neural network gives the heuristic. We introduce this nonuniform sampling strategy into RRT and RRT* and evaluate the algorithms on different kinds of maps. The simulation results demonstrate that the nonuniform sampling strategy significantly speeds up these algorithms and improves their stability.