Modeling Collision Avoidance Behavior With Zero-Speed Pedestrians

Echeverri, Laura C.; Auberlet, Jean-Michel; Hubert, Jean-Paul · 2024 · Crossref

DOI: 10.1109/tits.2024.3376077

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 a critical limitation in pedestrian crowd simulations: the occurrence of "livelocks" when agents interact with zero-speed pedestrians. Zero-speed pedestrians are defined as individuals who stop temporarily for activities such as resting or using a phone before resuming movement. While the Optimal Reciprocal Collision Avoidance (ORCA) algorithm is widely used for modeling collision avoidance, it was originally designed for agents in perpetual motion. Consequently, when a moving agent approaches a zero-speed agent in its path, ORCA can cause the moving agent to become blocked despite having available space to maneuver. This livelock invalidates simulation runs, hindering applications in crowd management, urban planning, and safety analysis. The authors aim to identify the causes of these livelocks and propose a modification to ORCA that prevents them without altering the algorithm’s general behavior or requiring auxiliary path-planning methods. The study employs a theoretical analysis and simulation-based experimental design to diagnose and resolve the issue. The authors first classify collision avoidance algorithms into force-based, velocity-based, vision-based, and data-driven categories, positioning ORCA within the velocity-based class. They analyze the mechanics of ORCA, which calculates collision-free velocities by intersecting half-planes of permitted velocities for each agent. To demonstrate the problem, the authors construct two specific simulation scenarios involving agents navigating a 10 × 5 m sidewalk with zero-speed obstacles. Using standard ORCA parameters (agent radius 0.3 m, maximum speed 1 m/s, time horizon 1 s), they illustrate how the algorithm’s reliance on minimizing deviation from a preferred velocity leads to agents slowing down indefinitely and becoming stuck behind stationary peers. The proposed solution modifies the ORCA logic to detect impending livelocks and compute a new collision-free velocity that allows the agent to change direction and bypass the zero-speed agent. The results demonstrate that the modified ORCA algorithm successfully prevents livelocks in the identified scenarios. In the original ORCA implementation, agents approaching zero-speed pedestrians experienced a continuous decrease in velocity magnitude until they were effectively immobilized, as the algorithm prioritized maintaining direction over finding alternative paths. The proposed modification allows agents to recognize when a livelock is likely and adjust their trajectory to navigate around the stationary agent. This ensures that simulations continue without interruption, maintaining realistic crowd flow dynamics. The authors validate this through illustrative examples showing that the modified algorithm enables agents to reach their destinations even when obstructed by temporary stops, whereas the unmodified algorithm fails. The significance of this work lies in its contribution to the robustness of agent-based pedestrian simulations. By addressing a specific gap in existing literature—livelocks caused by zero-speed agents rather than static obstacles or deadlocks—the authors provide a more accurate tool for evaluating crowd behavior in environments like shopping centers and public gatherings. The modification enhances ORCA’s applicability without increasing computational complexity or relying on external algorithms, making it a practical improvement for researchers and practitioners in intelligent transportation systems and crowd management. This advancement allows for more reliable predictions of pedestrian dynamics in scenarios where temporary stopping behaviors are prevalent.

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.

• pedestrian behavior perception