Assured Autonomy Using Dynamic Monitors and Simulation (ADAM DMS)

Systems that are forced out of the certified operating configuration can, and do, fail catastrophically. ln many cases, the systems can be saved by executing often counter intuitive actions. With a complex system, it is unlikely that a human operator would be able to "discover" these recovery actions before the system is destroyed. Using a high precision simulator, machine learning enables the research to learn action policies that can save a system from catastrophic failure, if it is caught quickly enough. The Assured Autonomy using Dynamic Monitors and Simulation (ADAM DMS) system learns the safe operating configuration and deep learning allows synthesis of a monitor and the learned recovery policy can be synthesized to perform the recovery. A simulated quadcopter was used as a demonstration example.