Autonomous Reconnaissance Trajectory Guidance at Small Near-Earth Asteroids via Reinforcement Learning

What is it about?

The spacecraft's ability to autonomously explore asteroids could reduce the cost and allow us to send multiple cost-effective spacecraft to various targets. A challenge in asteroid exploration is the lack of prior information about target asteroids, making it difficult for engineers and scientists to make a detailed flight plan ahead of time. Thus, an autonomous spacecraft needs to make such plans after arrival. Also, because the spacecraft's control has noise, the spacecraft needs to be able to modify its action according to the current status. As an approach to implementing such an adaptive planning capability, we study the application of reinforcement learning to design a hovering trajectory controller for surface global imaging tasks. We also perform high-fidelity numerical simulations, combining the proposed control strategy with a previously proposed onboard navigation scheme. The simulations show the proposed autonomous exploration approach is robust.

Featured Image