Description
The component of pedestrian-dynamics that handles short-range, reactive avoidance of contact between walkers — the local kinematics that turn a graph of intended trajectories into one without overlaps. Collision avoidance is what gives Social Force, RVO, and ORCA their characteristic micro-behavior; calibrating it correctly is what gives a crowd-simulation qualitative realism at moderate densities. It is also what fails first in crush scenarios.
Why it's hard
- Smartphone-distracted walking degrades anticipation and shifts collision-avoidance regime.
- Group cohesion (couples, families) overrides single-agent collision avoidance.
- Asymmetric avoidance — wheelchair users, elderly walkers — is rarely modeled.
- High densities (>4 ped/m²) push walkers out of the avoidance regime into the contact regime where models break.
- Empirical calibration requires high-resolution trajectories that are hard to collect.
Common approaches
- Social Force model with calibrated repulsive potentials.
- Velocity-Obstacle / RVO / ORCA reciprocal collision-avoidance schemes.
- Coupled mechanical-decisional models for distracted walking.
- Heuristic anticipation models (constant-velocity prediction with reaction).
Source Papers
- helbing1995_149d ↗ — Social Force model (foundational).
- echeverrahuarte2023_fcc4 ↗ — coupled mechanics + decisions for distracted pedestrians.
- zhong2022_7cb2 ↗ — data-driven crowd-modeling survey (collision avoidance covered).
- yang2020_e295 ↗ — review on crowd simulation and modeling.
- feng2021_f56d ↗ — data collection methods for studying pedestrian behaviour.