Toyota Sienna Incident

The May vehicle was involved in a rear-end collision shortly after completing a right-hand turn. As the vehicle proceeded through its turn, it encountered exhaust fumes from vehicles waiting at a stoplight in the adjacent lane. Due to environmental conditions and sensor limitations, the vehicle's lidar system interpreted the exhaust plume as a solid object in its path. In response, the autonomous vehicle first tried to adjust its position. When this maneuver proved insufficient, the system initiated a hard stop response. The vehicle behind, which was following at approximately 7-8 meters, was unable to stop in time and collided with the rear of the autonomous vehicle. The collision occurred at relatively low speeds (the autonomous vehicle had decelerated from approximately 7.6 m/s to 0 m/s), though exact impact speeds of the following vehicle were not recorded. The primary factor that led to this incident is that Lidar does not pierce through exhaust so the exhaust cloud appears to be a large, solid object. Some corrective actions that will be implemented: consider weather implication in future platform sensor selection and continue working on the software to improve a neural network to classify lidar points, which will use temporal information, radar, and camera features to make a decision on whether a lidar point is truly an object.