A Safety Reasoning Copilot Design for Future Uncrewed Ground Vehicles (UGVs)

Autonomous Navigation (AN) in complex-heterogeneous environments is an unsolved issue for both commercial and defense Autonomous Vehicle (AV) applications: A) Based on accumulated data through 2021 there are on average 9.1 driverless car crashes per million miles driven compared to 4.1 human-driven car crashes. B)The US Army recently reduced the requirement for its current Bradley replacement program of record from an “optionally manned fighting vehicle” to a system that “will not be something you operate entirely unmanned in its initial configuration”. C) Between 2021 and 2023 Ford, UBER, Lyft and Tesla have limited their fully AV operations due to safety related business concerns.

It is clear a research breakthrough is needed to ensure AV software is mature to a point where it can handle complex driving scenarios. In complex dynamic domains (e.g. intersections or congested terrain) the expected mode of operation for ensured safety of these unmanned systems is still direct human control (whether through direct vehicle input or teleoperation). This paper summarizes results of a completed PHD focused on the creation of an autonomous vehicle safety reasoning copiloting system that works.