Learning Energy-Efficient Trajectory Planning for Robotic Manipulators Using Bayesian Optimization

Abstract

Energy-optimal operation of robotic systems has gained high interest in both industry and science. We propose to fuse model predictive control and Bayesian optimization to plan minimum-energy trajectories for industrial robots that guarantee successful executions of the primary task. Particularly, parts of the predictive planner are learned using Bayesian optimization to account for the secondary, higher-level objective - here energy minimization. The effectiveness of the proposed approach is underlined in simulation, where a reduction in energy consumption is observed while maintaining a high quality of task executions.