Intelligent Motion Planning for Collision Free Autonomous Docking of Satellite Emulation Platform Using Reinforcement Learning

摘要

A reinforcement learning (RL) enabled intelligent motion planning for collision-free autonomous docking manoeuvre explicitly designed for a robotic floating satellite emulation platform is presented in this article. The Twin Delayed Deep Deterministic Policy Gradient-based RL algorithm involving deep neural network architecture in the actor-critic framework is considered to obtain the collision-free safe docking policy. The RL agents have been trained to perform a resilient target acquisition, ensuring its terminal position and velocity requirements while enabling the capability to avoid both static and dynamic obstacles. The resulting optimal policy is implemented as a feedback control law to enable computationally efficient onboard reactive motion planning for autonomous safe docking of the robotic floating satellite platform in a complex dense debris environment. The efficacy of the proposed motion planning scheme is validated with numerous simulation studies, where it is depicted that the trained RL-based planner has the potential to address the target acquisition with a sufficient degree of accuracy in the presence of both static and dynamic obstacle scenarios.