SUMMARY

Assured autonomy is increasingly critical to current and future DoD mission needs as the number of autonomous system designs continues to proliferate. While machine learning (ML) technology has advanced rapidly, formal safety assurances for these autonomous systems is lagging. This is largely due to their reliance on data-driven ML technologies which are inherently unpredictable and lack the necessary mathematical framework to provide guarantees on correctness. Without assurances, trust in any learning enabled cyber physical system’s safety and correct operation is limited, impeding their broad deployment and adoption for defense and other critical operating environments.

Patriot Labs is interested in the development of rigorous design and analysis technologies for continual assurance of learning-enabled autonomous systems, in order to guarantee safety properties and functional correctness in adversarial and/or unpredictable environments. Assurance must extend beyond the point of provisioning and throughout the time of operation.

For purposes of this CFI, solutions should incorporate some form of learning-enablement capable of operationalizing “background knowledge” that has been acquired and updated through a “learning process” – all while operating in dynamic and unstructured environments. Capabilities should include the continual monitoring, updating, and evaluation of system behaviors as changes within the environment and/or conditions occur.

Potential approaches may include supervisory learning for training classifiers, reinforcement learning for developing control policies, algorithms for learning system dynamics, or other methods for instantiating knowledge. Special consideration given to solutions that (i) do not rely on substantial operator involvement thereby reducing human-machine interaction and mixed initiative control requirements; (ii) deliver sustained high levels of autonomy in uncertain, unstructured, and dynamic environments; and (iii) apply predictable techniques and/or methods based on an understandable mathematical framework for providing guarantees on correctness.