Sampling-based Kinodynamic Motion Planning Using a Neural Network Controller
Dongliang Zheng, Panagiotis Tsiotras · AIAA Scitech 2021 Forum · 2021
View Video Presentation: https://doi.org/10.2514/6.2021-1754.vid This paper proposes a sampling-based kinodynamic motion planning method with several learning components. Kinodynamic RRT* for general nonlinear systems is computational inhibitive because many two-point boundary value problems (TPBVPs) are required to be solved repeatedly. This paper exploits supervised learning and offline optimal trajectories to build a state feedback neural network controller. The neural network controller aims to steers the system from points in an initial state set to points in a goal state set, thus avoiding the need for online solving TPBVPs. We develop a learning-based kinodynamic RRT* algorithm that is explicitly designed to incorporate the learned Neural Network controller and a cost-to-go Neural Network. Finally, Multiple simulation cases are studied to evaluate the performance of the proposed method.