Revealing Right-Turn Behavior of Human Drivers as a Model for Autonomous Vehicles 2021-01-0866

Although great progress has been made to improve the safety and performance of autonomous vehicles with the ultimate goal of meeting the public expectation of preventing most accidents, the current fleet of autonomous vehicles being tested continues to demonstrate that we still remain distant from that holy grail. One rationalization for some of these accidents has been that different maneuvers performed by such cars are not human-like (i.e. they do not display certain driving patterns to which human drivers are accustomed to). With that in mind, it would be hard to dispute the need for such vehicles to adapt to and somewhat imitate human driving in order to gradually integrate human-driven traffic in the future. In previously published work, we had examined human driver behavior when approaching and proceeding through stop-sign-controlled intersections using data obtained from a large-scale, on-road eye tracking study conducted in instrumented test vehicles to understand and assess human behavior in a naturalistic driving environment. We provided various metrics based on the vehicle dynamics to quantify human behavior at such intersections and suggested the use of such data for bettering autonomous vehicle performance. Here, we apply a similar approach to the analysis of driver behavior during right-turn maneuvers (specifically at stoplights and uncontrolled intersections). We obtained deceleration rates, stopping/slowing speeds, acceleration rates and angular velocities while participants drove specific routes in Los Angeles. As with our previous findings, we propose that our current data can be incorporated into autonomous systems to allow them to act more human-like (e.g. to avoid being rear-ended by other vehicles during right turns) and to better mimic human driving behaviors to provide smoother rides to passengers (e.g. minimizing abrupt changes in vehicle motion).