SBIR-STTR Award

This program will develop an autonomy stack for advanced air mobility using an innovative combination of formal-methods for design assurance and run-time-assurance to bound complex algorithms that are not suitable for design-time assurance. We decompose the aviate-navigate-communicate functions into control, perception, cognition, and decision-making modules for advanced air mobility aircraft. Analysis of delegation, supervision, and the overhead associated with human-autonomy teamwork is an inherent part of our formal design-time assurance. By incrementally replacing human-provided functions with the ability to delegate tasks to our assured, trustable autonomy stack, we provide the Air Force with a pragmatic path to certifiable simplified vehicle operations and beyond. The stack will be evolved systematically from Phases I to III from simulation, unmanned aircraft, simplified vehicle operations (Pipistrel, Skyborg), Air Force resupply (Phase III), Commercial Cargo Transport on Magnix Electric Caravan and eventually to a certifiable autonomy stack for commercial passenger operations. Critically, our approach addresses human-autonomy teamwork and autonomy certification (military and civilian) right from Phase I and apply it to multiple air-frames.

The program’s objective is to deliver the assured autonomy capabilities of Geofencing and Collision Safety to our customer, the Air Force Research Labs (AFRL), and our end-user, the Department of Defense Test Resource Management Center (TRMC). For example, the renowned Auto GCAS deals with the single significant problem of avoiding a ground collision. However, automatic geofencing and air-to-air collision safety between aircraft during formation flight and tactical rejoin maneuvers have yet to be accomplished and fielded operationally. During Phase II, Nodein and the California Institute of Technology will conduct the research and development needed to adapt Nodein’s commercial Assured Flight Autonomy Stack (AFAS) for AFRL and TRMC. Work will focus on three primary activities: Research and Develop in simulation the core Nodein-Caltech technologies for geofencing and air-to-air collision avoidance for end-user needs. Work with AFRL to assure the safety of reinforcement-learning-based tactical rejoin trajectories. Work with the US Air Force Test Pilot School at Edwards Air Force Base to test a sequence of experiments on manned jet aircraft in September 2022. The technologies developed will be flight test validated with the following objectives: Successfully fly automatic collision avoidance between a live jet and a virtual leader jet. Successfully enforce automatic geofencing on a live jet aircraft in the presence of commands that violate geofencing. Successfully fly automatic collision avoidance between a live jet and a live leader jet. Successfully fly simultaneous collision avoidance and geofencing safety constraints between two live jets. Collect data from a neural network trained without collision avoidance run-time-assurance. Collect data from a neural network trained with collision avoidance run-time-assurance.