Advanced Verification Toolset for Learning-Based UAS Operating in Uncertain Environments

This effort proposes to develop a toolset to evaluate the effectiveness of a safety-controller that monitors a learning-based autonomous control system. The traditional use of deterministic verification techniques for real-time monitoring is not feasible in the presence of uncertainty. This effort would advance the recommended verification techniques to include non-deterministic approaches. The output would be to develop a Mathworks-based toolset to evaluate the run-time safety controller using advanced verification techniques. The inputs to the tool would allow for unexpected / unknown inputs into the system under test to better reflect real world conditions. The effort increases the fidelity of the verification technique and evaluates a learning-based trajectory planner with a safety controller currently used to operate a helicopter. This effort examines the use of boundary certificates and other approaches to further improve the detection of issues that traverse a safety boundary.