Autonomous space systems naturally operate in partly unknown environments, presenting a major challenge to the presently available methods for mission and motion planning. While currently existing strategies do allow systems to adapt their plans in line with the results observed during the mission execution, capabilities of such approaches are limited. Namely, they do not incorporate heterogeneous information available from previous or parallel missions into the learning process, nor approach system safety specifications or unexpected situations in a formal way. Instead, the onus is on a human supervisor to deal with any issues in an ad hoc manner. This talk will present an overview and preliminary advances of the recently launched NASA Early Stage Innovations (ESI) project aiming to develop autonomous learning and control policies which guarantee system safety during learning and expedite learning by exploiting available prior and side information, while ensuring system resilience against a wide class of unexpected adverse scenarios.
