Towards learning-based trajectory tracking control for a planetary exploration rover: Adaptive model predictive control

Automous Robotic vehicles are important for exploring distant planets due to the risks associated with manned missions. However, they face challenges in navigating unknown terrain, requiring a deep understanding of wheel-ground interaction for effective slip estimation and compensation. Conventional methods have proven insufficient in addressing these challenges. The DeLeMIS project showcases AI methods to enhance autonomy in uncertain terrain, focusing on the trajectory tracking performance. This paper presents the development and evaluation of an adaptive model predictive control algorithm within the project, extended by machine-learning techniques. The learning-based algorithm and conventional baseline controllers are implemented and systematically evaluated through a comprehensive test campaign, which is described in our previous publication. This paper gives an overview of the control architecture and the training process, providing detailed insights into the neural network architecture and the results on a prototypic rover hardware.