Sharing the Road: How Human Drivers Interact with Autonomous Vehicles on Highways

Abstract

With more autonomous vehicles (AVs) being tested or deployed on public roads, human-driven vehicles (HVs) have to share the road with AVs. However, human drivers may not interact AVs the same way as they interact with HVs. Very few studies have investigated drivers’ behaviors when sharing the road with AVs. Based on a real-world dataset, our study explored drivers’ interactions with AVs in two types of events on highway, i.e., car-following event and car-passing event. The results show that, compared to interacting with HVs, drivers tended to keep a larger safety margin (i.e., larger gap distance and time gap) at high speed in both types of events when interacting with AVs. At the same time, drivers seemed to have difficulty anticipating AVs’ speed changes at high speed, as indicated by a larger standard deviation of the HVs’ speed and a smaller time to collision when following AVs versus following HVs.

Characteristics of Rear-End Collisions: A Comparison between Automated Driving System-Involved Crashes and Advanced Driving Assistance System-Involved Crashes

“…It seems that when proceeding straight, human drivers were less capable of anticipating the motion of an ADS-controlled vehicle compared with when there was lateral movement of human-driven vehicles. Previous research has observed a smaller time-to-collision when human drivers were following an ADS-controlled vehicle in mixed traffic compared with when following a human-driven vehicle (12). It might be possible that, when proceeding straight, the front view of the following driver is more likely to be blocked by the lead vehicle; if the ADS-controlled vehicle is not pre-active to the incidents ahead, the time budget left for the following vehicle to respond will be limited.…”

“…Specifically, from the start of the DMV program (2014) to Nov. 2018, rear-end collisions were 61.1% of all ADSinvolved crashes in the DMV dataset (8). Similarly, research based on the Waymo open dataset also revealed that, compared with human driven vehicles, ADS-driven vehicles led to smaller time-to-collision of following vehicles (12,22). The research by Song et al provided further details of ADS-involved rear-end crashes-it was found that the most representative temporal pattern of ADSinvolved crashes was ''collision following autonomous vehicle stopping'' (11).…”

“…It was found that the ADS being tested on roads at the current stage generally exhibited a worse safety record compared with human drivers (4)(5)(6)(7). Specifically, research found that rear-end crashes dominate the ADSinvolved crashes, possibly because of the reduced behavioral predictability of ADS-controlled vehicles (5,(8)(9)(10)(11)(12). The crash pattern of ADAS-controlled vehicles, however, has been under-investigated, possibly because of the lack of data.…”