Spacecraft rendezvous guidance in cluttered environments via artificial potential functions and reinforcement learning

The primary contribution of this work is to use Artificial Potential Functions (APF) for generating trajectories to be used as initial guesses for General Pseudospectral Optimal Control (or GPOPS). This work demonstrates dramatic speed up for GPOPS solution times, giving an average trajectory generation time of around 6 seconds. With this level of performance, the trajectory generation could occur on board the spacecraft based off of its current state estimate. In the type of scenarios that this algorithm is designed for (rendezvous, orbital transfer), this work can execute in near-real-time. This worked also improves the trajectory tracking controller performance, achieving continuous thrust fuel efficiency equal to the GPOPS optimal solution, and pulsed thrust fuel efficiency about 25% worse than the GPOPS optimal solution.