Obstacle-aided Locomotion for Underwater Snake Robot using Monte Carlo Model Predictive Control and Curvature Derivative Control

摘要

This paper studies a sample-based model predictive control using the Monte Carlo sampling method. Different from gradient-based Model Predictive Control (MPC), Monte Carlo Model Predictive Control (MCMPC) is able to introduce discontinuous phenomena such as collisions into the system. This feature makes it suitable for the control of underwater snake robots to explore complex and narrow spaces and interact with the environment when operating rescue missions. Curvature Derivative Control (CDC) is utilized to reduce the control inputs from each joint to only the head joint which optimizes the lateral undulatory locomotion of the snake robot and mitigates the complexity of MCMPC calculation. Collision is introduced into the system as a Linear Complementarity Problem (LCP), which allows MCMPC to respond to collisions in the prediction stage. Simulation results showed that the snake robot can generate various gait patterns and accomplish obstacle-aided locomotion by taking obstacles into account. The robot can reach the target area by adding positional error between the center of mass of the snake robot and the destination into the cost function.