[Paper Review] Safe, efficient and socially-compatible decision of automated vehicles: a case study of unsignalized intersection driving

Daofei Li, Ao Liu+2 more|arXiv (Cornell University)|Nov 4, 2021

Autonomous Vehicle Technology and Safety4 citations

TL;DR

This paper proposes a game-theoretic decision algorithm for automated vehicles (AVs) to safely and efficiently navigate unsignalized intersections while respecting social norms during interactions with human-driven trucks. By modeling truck drivers' limited visual fields and incorporating social fitness and reciprocal altruism into payoff design, the algorithm improves safety, efficiency, and human-driver expectations in human-in-the-loop experiments with 24 subjects and 207 interaction cases.

ABSTRACT

Safe and smooth interacting with other vehicles is one of the ultimate goals of driving automation. However, recent reports of demonstrative deployments of automated vehicles (AVs) indicate that AVs are still difficult to meet the expectation of other interacting drivers, which leads to several AV accidents involving human-driven vehicles (HVs). This is most likely due to the lack of understanding about the dynamic interaction process, especially about the human drivers. By investigating the causes of 4,300 video clips of traffic accidents, we find that the limited dynamic visual field of drivers is one leading factor in inter-vehicle interaction accidents, especially in those involving trucks. A game-theoretic decision algorithm considering social compatibility is proposed to handle the interaction with a human-driven truck at an unsignalized intersection. Starting from a probabilistic model for the visual field characteristics of truck drivers, social fitness and reciprocal altruism in the decision are incorporated in the game payoff design. Human-in-the-loop experiments are carried out, in which 24 subjects are invited to drive and interact with AVs deployed with the proposed algorithm and two comparison algorithms. Totally 207 cases of intersection interactions are obtained and analyzed, which shows that the proposed decision-making algorithm can not only improve both safety and time efficiency, but also make AV decisions more in line with the expectation of interacting human drivers. These findings can help inform the design of automated driving decision algorithms, to ensure that AVs can be safely and efficiently integrated into the human-dominated traffic.

Motivation & Objective

• To address the challenge of AVs failing to meet human drivers' expectations during unsignalized intersection interactions.
• To investigate how limited visual fields of truck drivers contribute to inter-vehicle accidents, especially involving AVs.
• To design a decision-making algorithm that enhances safety, efficiency, and social compatibility in AV-HV interactions.
• To evaluate whether incorporating social fitness and reciprocal altruism improves human drivers' perception of AV behavior.
• To provide empirical validation through human-in-the-loop experiments with real drivers interacting with AVs.

Proposed method

• A probabilistic model is developed to represent the dynamic visual field limitations of truck drivers during intersection interactions.
• A game-theoretic framework is designed where AV decision-making considers social compatibility through payoff functions incorporating social fitness and reciprocal altruism.
• The algorithm uses real-time perception and prediction of human-driven truck trajectories to inform strategic AV decisions.
• Payoff design integrates both safety (collision avoidance) and social factors (yielding behavior, turn-taking expectations) to guide AV behavior.
• Human-in-the-loop experiments simulate 207 unsignalized intersection interactions with 24 participants driving in a driving simulator.
• Three algorithms—proposed, baseline, and conservative—were compared to assess performance across safety, efficiency, and social compatibility metrics.

Experimental results

Research questions

• RQ1 How do limited visual fields of truck drivers contribute to accidents during unsignalized intersection interactions?
• RQ2 To what extent does incorporating social fitness and reciprocal altruism improve AV decision-making in human-in-the-loop settings?
• RQ3 Can a game-theoretic AV decision algorithm achieve better safety and efficiency while aligning with human drivers' expectations?
• RQ4 How do human drivers perceive and respond to AV behavior that reflects social compatibility versus purely safety- or efficiency-driven strategies?
• RQ5 What is the relative impact of visual field modeling on AV decision performance in complex intersection scenarios?

Key findings

• The proposed algorithm significantly improved safety by reducing collision risk during unsignalized intersection interactions.
• Time efficiency was enhanced compared to baseline and conservative algorithms, with faster and more predictable decision-making.
• Human drivers reported higher acceptance and perceived naturalness of AV behavior when social compatibility was incorporated.
• The algorithm outperformed alternatives in aligning with human expectations, particularly in turn-taking and yielding behavior at intersections.
• The integration of visual field modeling improved the realism and accuracy of AV predictions about human truck drivers.
• The human-in-the-loop experiments confirmed that social compatibility is a critical factor in AV-HV interaction success.