Hazard Detection and Avoidance Systems for Autonomous Planetary Landing

Abstract

Precise and autonomous landing capability is a key feature for the next space systems generation. The possibility to carry out Hazard Detection and Avoidance would allow both absolute and relative correction maneuvers, dramatically increasing the robustness and the flexibility of the mission. A novel guidance algorithm is presented: the trajectory is modeled as a polynomial of minimum degree required to satisfy the boundary constraints, leaving a reduced set of parameter free to be optimized. The novelty of the proposed method lies in the fact that it allows the computation of large diversions, with a suboptimization of the fuel consumption, satisfying at the same time all the constraints imposed by the system in terms of control torques, thrust, and allowed hovering area. Only 2 or 3 optimization variables are needed, making the algorithm light enough to run on-board. The flexibility of the guidance is addressed with two different applications, a lunar landing and the close approach to an asteroid. An ad hoc optimizer is also developed, based on Differential Algebra, capable to solve the guidance optimization in a fast and reliable way. Objective and constraints are modeled as low order Taylor maps. The general features of the functions are easily got, leading in a few iterations to the optimal solution, due to the property of the Taylor series to converge to the true value in proximity to the expansion point. An innovative hazard detection and target selection algorithm is also proposed. The capability of Artificial Neural Networks (ANNs) to extrapolate underlying rules in complex datasets is exploited to obtain an automatic classifier that builds a hazard map of the landing area, basing on a single image. Manual establishment of heuristic correlations between image and terrain features is no longer required, leaving to the ANN training the task to identify these correlations automatically; the process is run off-line, while only the trained network runs on-board, with a minimal computational burden. A target selection algorithm exploits the map to locate and rank the candidate landing sites following safety and reachability criteria. A coherent and effective dataset for rigorous training and test is generated with a realistic simulation tool. The network showed the ability to select a safe landing site in 100 % of cases.

Paolo Lunghi

Assistant Professor of Aerospace Systems

Aiming for autonomous Guidance, Navigation, and Control for spacecraft.