Iterative learning-based model predictive control for mobile robots in space applications

This paper presents an iterative learning-based model predictive controller (MPC) for trajectory tracking control of an autonomous planetary rover on unknown terrain. In order to achieve accurate trajectory tracking under model uncertainties, a nonlinear controller and an MPC are utilized, combined with a learning-based uncertainties approximation. The model uncertainties and disturbances are learned using a deep neural network (DNN) as well as a parametric model and results are compared. For test and validation purposes, a gazebo simulation is used, which is itself already validated using data from a prototype rover. With that, the trajectory tracking performance of the proposed learning-based MPC is validated and compared to other well-performing controllers. The results show that the algorithm is able to learn model uncertainties and to compensate them during runtime while being practicable for the implementation and in the training phase.