Imitation Learning for Autonomous Trajectory Learning of Robot Arms in\n Space

Abstract

This work adds on to the on-going efforts to provide more autonomy to space\nrobots. Here the concept of programming by demonstration or imitation learning\nis used for trajectory planning of manipulators mounted on small spacecraft.\nFor greater autonomy in future space missions and minimal human intervention\nthrough ground control, a robot arm having 7-Degrees of Freedom (DoF) is\nenvisaged for carrying out multiple tasks like debris removal, on-orbit\nservicing and assembly. Since actual hardware implementation of microgravity\nenvironment is extremely expensive, the demonstration data for trajectory\nlearning is generated using a model predictive controller (MPC) in a physics\nbased simulator. The data is then encoded compactly by Probabilistic Movement\nPrimitives (ProMPs). This offline trajectory learning allows faster\nreproductions and also avoids any computationally expensive optimizations after\ndeployment in a space environment. It is shown that the probabilistic\ndistribution can be used to generate trajectories to previously unseen\nsituations by conditioning the distribution. The motion of the robot (or\nmanipulator) arm induces reaction forces on the spacecraft hub and hence its\nattitude changes prompting the Attitude Determination and Control System (ADCS)\nto take large corrective action that drains energy out of the system. By having\na robot arm with redundant DoF helps in finding several possible trajectories\nfrom the same start to the same target. This allows the ProMP trajectory\ngenerator to sample out the trajectory which is obstacle free as well as having\nminimal attitudinal disturbances thereby reducing the load on ADCS.\n