Fig. 4: Performance evaluation for a real-world AV at a physical test track.
From: Breaking through safety performance stagnation in autonomous vehicles with dense learning
a Illustration of the real autonomous vehicle (AV) under test, equipped with Autoware, Lidar, cameras, on-board computer, by-wire controller, high-definition (HD) map, and RTK (Real-Time Kinematics) GPS (Global Positioning System). b Illustration of the Mcity test track including highways, roundabouts, intersections, urban streets, etc. c Illustration of the mixed-reality environment combining the physical road infrastructures, proxy physical objects, and a simulation environment, where information of the real world and simulation world is synchronized. d Safety performances of SafeDriver in the co-simulation of SUMO and Autoware at Mcity. e Field testing results of the real AV with SafeDriver regarding the overall crash rate, crash rates of different crash types, and the avoidable crash rate. f Cases of SafeDriver for avoiding crashes in safety-critical situations. In the first case, the SafeDriver (red vehicle) made emergency braking with right steering to avoid collisions in the situation that the background vehicle in the right lane made a reckless cut-in, while the vehicle from the opposite direction was approaching. In the second case, the SafeDriver (red vehicle) made emergency braking with left steering to avoid collisions in the situation that a background vehicle failed to yield when entering the roundabout. Additional explanations are available in Supplementary Videos 7−8.
