Decision-making for automated vehicles at intersections adapting human-like behavior

de Beaucorps, Pierre; Streubel, Thomas; Verroust-Blondet, Anne; Nashashibi, Fawzi; Bradai, Benazouz; Resende, Paulo · 2017 · OpenAlex-citations

DOI: 10.1109/ivs.2017.7995722

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the challenge of developing naturalistic decision-making algorithms for automated vehicles (AVs) at unsignalized intersections and roundabouts. While automated driving systems prioritize safety, they often lack the adaptability and robustness of human drivers, leading to overly conservative behavior. The authors propose a method that combines human-like driving strategies, derived from recorded driver data, with an objective risk assessment to ensure collision avoidance. The goal is to create an AV that can safely decide whether to pass before or yield to another vehicle, mimicking human comfort limits while maintaining strict safety boundaries. To achieve this, the researchers developed a custom multi-agent simulation environment capable of processing human inputs in a controlled setting. The simulator uses a realistic physics engine and allows for the simultaneous driving of human participants and automated agents. Data was acquired from ten licensed drivers who navigated 56 setups across seven scenarios, including various crossroad maneuvers (left, right, and straight turns) and roundabouts. The human drivers’ speed profiles were recorded and clustered using k-means algorithms to generate reference speed profiles representing "passing first," "yielding," and "stopping" behaviors. These profiles serve as the basis for the AV’s trajectory planning, which separates path planning (offline) from speed profile assignment (online). The decision-making algorithm utilizes Post-Encroachment Time (PET)—the time gap between vehicles exiting and entering a conflict zone—as the primary risk metric. As the AV approaches an intersection, it predicts short-term trajectories for itself and other vehicles. It iterates through the reference speed profiles to find candidates that satisfy a minimum PET threshold. If multiple safe profiles exist, the algorithm selects the one with the highest instantaneous speed to maximize time efficiency; if no safe profile exists, it defaults to stopping. The system was validated by testing two versions of the AV with different PET thresholds against the same scenarios used for data acquisition. The results demonstrate that the algorithm successfully navigated all setups without collisions. Compared to human drivers, who experienced collisions due to the artificially risky nature of the test scenarios, the AVs maintained safety while exhibiting varying degrees of conservatism. An AV with a higher risk tolerance (lower PET threshold) behaved similarly to "daring" human drivers, frequently passing before other vehicles when safe, whereas a more conservative AV yielded more often. The study concludes that combining extracted human speed profiles with objective risk assessment allows AVs to handle standard intersection scenarios safely and naturally, bridging the gap between rigid automated logic and human-like adaptability.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• driver vru interaction

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Theoretical Contribution: computational model