ESA GNC Conference Papers Repository

Autonomous, precise and safe landing capability is a key feature for the next space missions. Often, scientifically relevant places are associated with hazardous terrain features; in other cases there is no possibility to completely characterize a predefined landing area (e.g. presence of a dense atmosphere, or small bodies that are too difficult to be observed such as asteroids). The chance to adapt and correct the final landing pinpoint almost up to the touch down increases both the robustness and the flexibility of the vehicle operations. In this paper, an hazard detection system, based on neural networks, able to reconstruct an hazard map of the landing area from a single image during the descent, is presented. During algorithms development it is very difficult to consider in advance all the types ofmorphological structures that can be encountered; the neural network approach allows to obtain a flexible system, able to operate also in conditions not explicitly considered during the project. Different networks analyze images at different scales, extracting from graphical information a certain number ofindexes ascribable to physical properties, such as shadows, surface roughness and slopes. Then, these values are fused in a unique hazard map. Different network training methods are investigated: the use of both artificial and real images is compared. Results for different scenarios in a lunar landing case are shown and discussed, in order to highlight the effectiveness of the proposed system. Sensitivity to environmental parameters, such as light conditions, trajectory inclination and camera attitude is investigated. Finally, possible future improvements are suggested.