Modelica FMI based hybrid reinforcement learning enhanced trajectory planning for an ADR scenario for combined control of a satellite with a 7-axis robotic arm using Modelica/FMI

Abstract

This work describes a novel hybrid reinforcement learningenhanced trajectory planning algorithm for an active debrisremoval scenario for combined control of a satellite with a7-axis robotic arm. A reinforcement learning algorithm iscombined with a correction algorithm and classicaltrajectory planning to handle the collision free approachof a chaser satellite to a target, and placing the gripperat the robots near the grasping point for use with acombined controller, which commands the satellite and itsrobotic arm simultaneously.The algorithm is verified using a complex simulationscenario study implemented in Modelica/FMI.