Validation and Verification of Safety-Critical Perception and Optimal Control for Autonomy in Orbital Robotics

Kurzfassung

The validation and verification (V&V) of safety-critical functionalities of space robots engaged in autonomous task-execution is facing new challenges. The recognized advantage of optimal control and machine learning for providing planning and perception capabilities onboard a robotic spacecraft calls for the development of new V&V techniques to ensure robust, explainable, and re- silient autonomous behavior. This paper presents V&V techniques and developments at DLR of safety-critical aspects specifically for the autonomous robotic capture of a target satellite. The paper first addresses optimal control for robustly planning the robot arm's interception with a predefined grasping point on the tumbling target satellite, as well as for providing the means of performing mission planning in view of operational and motion constraints. Deep-learning based perception algorithms are then ad- dressed, underlining the advantage of these with respect to classical approaches and presenting new methods for their V&V.