Multi-Objective Spacecraft Trajectory Optimization with Synthetic Agent Oversight

Numerous techniques exist to optimize aircraft and spacecraft trajectories over cost functions that include terms such as fuel, time, and separation from obstacles. Relative weighting factors can dramatically alter solution characteristics, and engineers often must manually adjust either cost weights or the trajectory itself to obtain feasible solutions. This work integrates a rule-based planner inspired by human cognition with an optimal controls trajectory planner to automatically construct trajectories that do not require manual inspection or adjustment. The cognitive agent translates mission goals into cost function weights expected to produce motions that appropriately trade fuel and time efficiency as well as proximity to obstacles. The quality of the resulting full-state trajectory is then evaluated based on a set of computed trajectory features and specified constraints. Although each trajectory is mathematically optimal with respect to its dynamics and the weighted cost function, the agent may find it unacceptable locally (e.g., passes through an obstacle) or globally (e.g., requires too much fuel). The violating condition(s) are either translated to a new weight set or the trajectory is locally repaired, iterating until an acceptable trajectory is generated or the domain is deemed unsolvable. An ideal planar robot implementation introduces the models and provides intuitive baseline results. A three-dimensional spacecraft implementation is presented, a domain in which fuel savings and safety are critical for success.