How to evolve safe control strategies

Autonomous space vehicles need adaptive control strategies that can accommodate unanticipated environmental conditions. The evaluation of new strategies can often be done only by actually trying them out in the real physical environment. Consequently, a candidate control strategy must be deemed safe - i.e., it won't damage any systems - prior to being tested online. How to do this efficiently has been a challenging problem. We propose using evolutionary programming in conjunction with a formal verification technique (called model checking) to evolve candidate control strategies that are guaranteed to be safe for implementation and evaluation.