Smooth PRM Implementation for Autonomous Ground Vehicle

Abstract

Generating continuous and smooth paths with collision avoidance that avoid sharp turns is a significant challenge for autonomous mobile robot navigation. Sampling-based motion planners do widely use in robotics due to their computing efficiency, flexibility, and simplicity. One of the sampling-based planners, Probabilistic Roadmap(PRM), starts with a random sampling of the points in the free space. Although this sampling-based planner is generally very efficient, it can occasionally become computationally expensive when it runs dangerously close to an obstacle. In addition, the computed path can contain sharp turns challenging for the differential drive robot to navigate. Also, the path is not optimal and can be longer than necessary. The idea presented in this paper is to demonstrate how to use the gradient descent approach to find an optimal (smoother) path even though PRM provides a longer path with abrupt turns. PRM and Smoothened PRM were both run on the given operational environment and compared the performance in simulation and hardware. The simulation result shows that the algorithm can shorten the length of the searched path. The smoothness of the path has significantly improved even if the PRM offers a path with abrupt turns. Moreover, the proposed algorithm runs well on Turtlebot3 waffle pi, performing real-time obstacle avoidance.