Browse Publications Technical Papers 2019-01-1021

Learning from Human Naturalistic Driving Behavior at Stop Signs for Autonomous Vehicles 2019-01-1021

Despite public expectations that autonomous vehicles should be able to avoid most accidents, the existing fleet of autonomous test vehicles has demonstrated this is simply not the case. An explanation for some of these accidents has been that these vehicles do not drive like humans and therefore do not exhibit certain driving patterns expected by human drivers. With the high likelihood of a gradual integration of autonomous vehicles into our traffic system in the future, there will be a need for such vehicles to adapt to, and mimic, human driving. Although much work has been done to understand human behavior and performance in driving, it has been mostly geared towards defining human capabilities and limitations. Little work has been done on the interactions between human-driven and autonomous vehicles. In previously published work, we described a large-scale, on-road eye tracking study conducted in instrumented test vehicles to understand and assess human behavior in a naturalistic driving environment. Here we describe one condition from that study, approaching and proceeding through stop-sign-controlled intersections, in the context of applying our work to the development of autonomous vehicles. We investigated naturalistic driver behavior at stop signs based on vehicle dynamics. In particular, we obtained deceleration rates, stopping/slowing speeds, stopping/slowing durations and acceleration rates while participants drove specific routes in Los Angeles. We also found and quantified clear evidence of “California Rolls” for many drivers. We suggest that data such as that presented in this paper can be incorporated into autonomous systems such that they behave more like human drivers (e.g. to avoid rear-end accidents), as well as to better predict human driving behaviors (e.g. “California Rolls” and very short stops).


Subscribers can view annotate, and download all of SAE's content. Learn More »


Members save up to 18% off list price.
Login to see discount.
Special Offer: Download multiple Technical Papers each year? TechSelect is a cost-effective subscription option to select and download 12-100 full-text Technical Papers per year. Find more information here.
We also recommend:

Optimizing Connected Vehicle Content Strategy for Drivers


View Details


A Look-Ahead Driver Model for Autonomous Cruising on Highways


View Details


Driver characteristic using driving monitoring recorder


View Details