Phantom braking in automated vehicles: A theoretical outline and cycling simulator demonstration

Hegna Berge, Siri; de Winter, Joost; Feng, Yan; Hagenzieker, Marjan · 2024 · Crossref

DOI: 10.54941/ahfe1005212

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 safety implications of "phantom braking"—sudden, unexpected deceleration in automated vehicles (AVs)—specifically regarding its impact on vulnerable road users like cyclists. The authors conceptualize phantom braking through the lens of signal detection theory, framing it as a by-product of AV decision-making where systems prioritize avoiding accidents over minimizing false positives. This conservative approach, while reducing pedestrian collision risks, increases the frequency of unwarranted braking events, potentially causing rear-end collisions or falls for trailing cyclists. The study aims to explain this phenomenon and demonstrate its effects on cyclist behavior using a controlled simulation environment. To investigate this, the researchers employed a newly developed virtual reality (VR) cycling simulator at Delft University of Technology. The experimental design involved a single participant cycling behind an AV programmed with four distinct braking behaviors: "Inaccurate," "Conservative" (high phantom braking), "Liberal" (low phantom braking, high miss rate), and "Perfect." Each condition consisted of 16 pedestrian crossing scenarios, varying the presence of pedestrians and the AV’s response. The AV maintained a fixed distance from the cyclist, and data on speed and brake reaction times (BRTs) were collected and filtered to analyze the cyclist’s responses to the AV’s actions. The results revealed that the cyclist consistently mirrored the AV’s braking behavior. In all instances where the AV decelerated, whether for a genuine hazard or a phantom brake, the cyclist also braked, with reaction times ranging from 1.0 to 1.6 seconds. Conversely, when the AV failed to brake for a present pedestrian, the cyclist also maintained speed, effectively ignoring the pedestrian. This behavior indicates a significant complacency effect, where the cyclist relied on the AV’s implicit and explicit signals rather than independent assessment of the environment. The cyclist primarily looked ahead at the AV, only glancing at pedestrians when stopped or moving slowly. The study concludes that phantom braking is an inherent trade-off in current AV sensor systems, which often favor conservative thresholds to minimize liability and accident risks. The findings highlight a critical safety concern: cyclists may exhibit automation bias, disregarding their own legal obligations and environmental cues in favor of the AV’s actions. This reliance increases the risk of rear-end accidents during phantom braking events and failure to yield to pedestrians when the AV does not. The paper validates the use of VR cycling simulators for further research into AV-cyclist interactions and calls for larger-scale studies to explore these behavioral dynamics and their real-world safety implications.

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.

• automation surprise