Nonlinear Model Predictive Control Applied to Vision-Based Spacecraft Landing

Real-time optimal control has eluded practical implementation for most systems so far. The reason being mainly related to the scarce computational resources available and the high CPU requirements of commonly proposed real-time optimal control architectures. In this paper we show how, by a careful use of the Nonlinear Model Predictive Control approach one can obtain a real-time control system able to drive a mass optimal spacecraft landing in the presence of highly noisy navigation inputs such as those coming from a light weight solution including only one IMU and a camera. The introduced approach is applicable to a broader class of systems, as is shown by applying the method to find time-optimal maneuvers for a quad rotor model.