The Safety of Autonomous Ground Vehicles as a Function of Sensing Capability

Autonomous ground vehicles (AGVs) promise to solve problems in several areas-from surveillance to manufacturing to freight transportation. With the successful culmination of the DARPA Urban Challenge, and some researchers road-testing autonomous vehicles across the Eurasian landmass, there is significant hope for commercialization. However, there is a fundamental obstacle: how do we guarantee autonomous vehicles will not make the roads less safe than they are now? The key to resolving this issue lies in developing guarantees of safety for autonomous ground vehicles in multi-vehicle environments at different speeds, road conditions, and the sensing capability needed to guarantee such safety. We develop these requirements for a limited set of traffic situations--for vehicles on two way roads, four-way stop signs and intersections. Our requirements are based on the safety of human drivers-to keep the probability of accidents less than that for good human drivers. These requirements will apply to any sensing solution, based on the fusion of a few or several sensors. We conclude with some analysis of solutions based on lidar, radar and camera sensors.