Effectiveness of Automated Driving Systems Assessed Using Real-World Naturalistic Driving Data from 50 Participants in the Washington, DC Area over 12 Months.
Real-World Use of Automated Driving Systems and Their Safety Consequences: A Naturalistic Driving Data Analysis
Automated Driving Systems (ADS) have the potential to extend active safety systems by continually assisting drivers with steering and longitudinal control elements of dynamic driving tasks. While the use of ADS grows over time, how these systems and their capabilities are perceived by road users, and knowledge surrounding the real-world use and effects on transportation safety are largely unknown. By combining analyses of naturalistic driving data collected in the Washington, DC area with a post-study survey, researchers investigated the potential safety benefits and consequences of ADS.
The research team investigated an existing naturalistic driving database collected from the Virginia Connected Corridor Level 2 Naturalistic Driving Study with data from 50 drivers using vehicles with at least longitudinal control systems such as Adaptive Cruise Control (ACC). Most drivers also had various lateral control systems available, such as Lane Keep Assist (LKA) or Lane Centering Assist (LCA). The drivers primarily commuted in the Washington, DC metropolitan area, and provided data for 12 months totaling an estimated aggregate 684,931 miles driven. Vehicular data such as location, headway, speed, acceleration, and videos of the forward roadway and driver actions were recorded. A total of 235 safety-critical events (SCEs) from the data were extracted through the use of kinematic algorithms identifying events such as hard decelerations and lane departures, and then confirmed by manual verification by analysts. A post-study questionnaire was also conducted to capture drivers’ subjective ratings on the usefulness and usability of the ADS.
- Results showed that 47 (20 percent) out of the 235 identified SCEs involved ADS use. All of these 47 SCEs were categorized in the data as near-crashes.
- For the 47 SCEs that occurred with at least one ADS feature activated, researchers analyzed the nature of the ADS involvement and the driver's behavior. Overall, in 87 percent of these cases, researchers considered the ADS to have either helped avoid a collision or at least not negatively impact the situation.
- In 14 (29.8 percent) of these cases, the ADS controlled the vehicle or issued an alert to the driver.
- In 7 (14.9 percent) of cases, researchers determined that the ADS was not related to the direction of the threat (e.g., potential rear-end crash hazard while LKA engaged).
- Drivers manually disengaged the ADS in 20 cases (42.6 percent), all of which involved taking over ACC control, due to other vehicles attempting to cut in ahead of their vehicle.
- In the remaining 6 cases (12.8 percent), researchers classified the event as a silent failure, where the ADS was expected to prevent the SCE but neither alerted the driver nor avoided the SCE.
- An in-depth analysis of these 47 SCEs revealed that people operated the ADS outside of the intended use in 57 percent of SCEs. Unintended uses included drivers engaged in secondary tasks, driving with both hands off the wheel, using the systems in adverse weather, or using systems on roads other than highways.
- Based on the post-study survey, participants agreed that ADS were effective at preventing crashes (86 percent of participants responded positively to ACC and 42 percent responded positively to LKA), performed as they expected (93 percent for ACC and 65 percent for LKA), and were easy to use (95 percent for ACC and 81 percent for LKA). The survey results also revealed that participants felt more comfortable engaging in secondary tasks when ADS were active compared to when they were not active (49 percent for ACC and 35 percent for LKA).
- For situations where ADS did not work well or worked below participants’ expectations, the most frequently reported factor (23 percent of participants) was insufficient ADS braking for a cut-in ahead of the vehicle, and insufficient acceleration when traffic in front was clearing (14 percent of participants).