CLOSED-LOOP NEURAL CONTROL FOR THE FINAL PHASE OF A CISLUNAR RENDEZVOUS

The paper presents the design of a closed loop controller based on neural networks for the final rendezvous in a three body problem scenario with full six degrees of freedom dynamics. The reference scenario is a rendezvous maneuver between an active chaser and a passive target located on a near rectilinear halo orbit around the L2 Lagrangian in the Earth - Moon system. The guidance system synthesis is performed using ”imitation learning” with two neural networks for attitude and translation. The performance of the controllers is evaluated with respect to a standard PID approach, as well as a high frequency, high performance state dependent Riccati equation algorithm, in order to evaluate the size of the learning database and capabilities of the two networks.