Bayesian Optimization Based Self-Improving Sliding Mode Controller For a Bio-Inspired Marine Robot

In this study, we propose and validate a novel control methodology combining Sliding Mode Control (SMC) and Bayesian optimization for a bio-inspired marine robot. The proposed approach uses minimal motion knowledge, excluding detailed dynamic parameters, to perform online updates of the sliding surface. The robustness of SMC, combined with the adaptive parameter tuning capability of online optimization, allows the controller to maintain stability under Lyapunov conditions, thus enabling safe experimentation and online optimization. The experimental results demonstrate the effectiveness of this method, showing significant improvements in maintaining the target depth and pitch angle of the robot after 56 iterations of online parameter updates. Furthermore, the stability of the system is guaranteed even when performing controller tuning searches, allowing for safe and effective real-time optimization. This holistic approach will not only improve the robot’s maneuverability and stability but also contribute to advancing the field of bioinspired underwater robotics.