Do You Want Your Autonomous Car To Drive Like You? (1802.01636v1)

Abstract: With progress in enabling autonomous cars to drive safely on the road, it is time to start asking how they should be driving. A common answer is that they should be adopting their users' driving style. This makes the assumption that users want their autonomous cars to drive like they drive - aggressive drivers want aggressive cars, defensive drivers want defensive cars. In this paper, we put that assumption to the test. We find that users tend to prefer a significantly more defensive driving style than their own. Interestingly, they prefer the style they think is their own, even though their actual driving style tends to be more aggressive. We also find that preferences do depend on the specific driving scenario, opening the door for new ways of learning driving style preference.

• The paper finds users significantly prefer a safer autonomous driving style as passengers ($p = .0121$) compared to their own spontaneous driving habits, revealing a mismatch between perceived ($r = .86$) and actual ($r = .40$) styles.
• Using a driving simulator, researchers evaluated user preferences across scenarios, demonstrating that while people perceive their preferred style to be like their own, this perception doesn't match their actual driving behavior.
• Findings imply autonomous vehicle development should focus on user safety preferences rather than replicating driver habits, suggesting the need for alternative feedback methods beyond inverse reinforcement learning from demonstrations.

Investigating Autonomous Car Driving Style Preferences

The paper "Do You Want Your Autonomous Car To Drive Like You?" explores the assumption that users might prefer autonomous cars to emulate their own driving styles. This research challenges the notion that an aggressive driver would want an aggressive autonomous vehicle, or a defensive driver a defensive one. Instead, the paper reveals a discrepancy between perceived and actual driving preferences, indicating that users often favor a safer driving style when they are passengers, diverging from their inherent driving behaviors.

Methodology and Findings

The researchers employed a driving simulator to evaluate user preferences across distinct driving scenarios. Participants first demonstrated their driving styles by navigating a course designed to mirror real-world traffic conditions. Subsequently, they experienced autonomous vehicle operation under various styles: their own (unknown to them), aggressive, defensive, and a distractor (another user's style). The paper's main metrics included perceived similarity of the driving style to their own and preference for specific autonomous car driving styles.

Notably, participants typically preferred more defensive driving styles compared to their actual spontaneous driving. A matched pairs $t$-test showed a significant difference ($p = .0121$), supporting the hypothesis that people favor more cautious driving behavior as passengers than when they are behind the wheel themselves. Furthermore, while participants often selected the style they perceived to be closest to their own, this style was not accurately representative of their true driving behaviors. The paper found a compelling correlation ($r = .86$) between perceived own style and preferred autonomous driving style. Yet, a weaker correlation ($r = .40$) existed between perceived style and actual style, signifying a misalignment.

Implications for Autonomous Vehicle Development

The findings hold potential implications for autonomous vehicle manufacturers and developers. The research highlights the necessity to tailor autonomous vehicles to match what users truly want in terms of safety, rather than replicating their driving habits. This insight advocates the design of autonomous driving systems that adapt dynamically based on both user input and environmental context, turning towards customization that accommodates perceived preferences rather than demonstrated behaviors.

The requirement to extract preferred driving styles might necessitate novel approaches. The findings challenge traditional methods relying solely on inverse reinforcement learning from user driving demonstrations, indicating a need for systems that can capture users’ desired driving experiences through alternative feedback mechanisms.

Future Directions

The paper opens avenues for further exploration in automotive behavior learning techniques, potentially incorporating user feedback and predictive algorithms that align more closely with passenger expectations in various driving contexts. Additionally, expanding research to diverse driving conditions and incorporating physiological metrics or self-reported comfort levels could add layers to understanding user preferences comprehensively.

This work could lead to advancements in personalized autonomous driving modes, potentially reducing user anxiety and increasing acceptance of autonomous driving solutions. Researchers and developers may pursue integrating user values and psychological assessments in autonomous systems, thereby enhancing user satisfaction and safety perceptions.

In conclusion, this paper presents critical insights into user preferences regarding autonomous driving styles. It implies a shift in how user-centric driving behaviors should be understood, focusing on their aspirations rather than their automatic actions. Understanding the nuances between perceived and preferred driving styles could significantly influence the future development of autonomous technologies, fostering vehicles that genuinely resonate with user expectations.