Online motion planning and control for autonomous on-orbit assembly with machine learning-based model predictive control

The future of space exploration missions highly depends on autonomous space systems. To move toward autonomy, the development of online motion planners becomes a priority, where the Model Predictive Control framework is superior in this domain. At the same time, machine learning techniques open up new horizons for designing autonomous systems. This paper proposes a novel method integrating machine learning techniques with model predictive control to perform on-orbit assembly autonomously using a robotic spacecraft. In this work, a set of machine learning models, trained using the datasets obtained from the high-fidelity model developed in our previous research, are proposed to predict a set of optimization parameters in the large-scale optimal control problem to accelerate computations for online planning. More specifically, the proposed machine learning architecture is trained on a dataset generated from the shape, size, position, and orientation of space structures present in the assembly environment to predict the closest point on space structures with respect to others; hence the minimum distance. This approach reduces the number of optimization parameters in the respective optimal control problems and speeds up computations drastically. The resulting machine learning models are later utilized within a model predictive control’s prediction horizon to propose fast online motion planners. Numerical simulations demonstrate that the proposed architecture maintains a small generalization error in computing and tracking distance between objects. Finally, the online motion planners are applied to an autonomous on-orbit assembly operation using a robotic spacecraft to show the efficiency and capability of the proposed approach. • Online motion planning and control for autonomous on-orbit assembly. • Nonlinear model predictive control for online motion planning and control of robotic spacecraft. • Machine Learning incorporation into model predictive control for motion planning and control. • Contributing to proximity operations and space exploration technologies.