Nonlinear Model Predictive Control and Reinforcement Learning for Capsule-Type Robot with an Opposing Spring

The paper considers the solution to two control optimization problems for a capsule-type robot with an opposing spring. The first problem focuses on optimizing the average velocity of the robot while the other one deals with a time-optimal control problem. The capsule-type robot is an essentially nonlinear system that performs cyclic reciprocating motions to move in any direction. In this study, the usage of Reinforcement Learning and Nonlinear Model Predict approaches for such a nonlinear system is researched. The trajectories and control laws are obtained, and the results are compared with the optimization results for a simple periodic piece-wise constant law with one switch. The study was conducted by methods of mathematical modeling.