Physics-Enhanced Deep Learning for Instantaneous Prediction of Dynamic Behavior of a Space Vehicle-Robot System

The operation of dynamic systems such as robotic arms, drones, and satellites can be improved if operators can predict their behavior. The systems have complex underlying physics that can be represented by their equations of motion. Normally, predictions of system behavior must be made by solving the equations of motion for the system of interest to get the state trajectories for that system. However, solving equations of motion for complex systems can be time consuming, as these solution require the use of numerical solvers. For use in real systems, dynamics predictions must be made almost instantaneously. Deep learning models can be used to make these near-instant predictions. This paper uses three deep learning approaches to quickly and accurately determine the state trajectories of the system of interest: a deep neural network, a physics-guided neural network, and a physics-informed neural network. The system of interest for this paper is a five-degree-of-freedom space vehicle with an attached robotic arm. The results of this paper show that each deep learning approach achieves accurate, fast predictions of the space vehicle’s future states. This shows that deep learning can be used for behavior predictions in real-time systems.