Reproduce “Active inference as a unified model of collision avoidance behavior in human drivers” on OpenResearch

About this paper

Julian F. Schumann, Johan Engström, Leif Johnson, Matthew O'Kelly, Joao Messias, Jens Kober, Arkady Zgonnikov

A computational cognitive model based on active inference accurately reproduces human collision avoidance behavior, including response timing, maneuver selection, and collision outcomes across front-to-rear, lateral incursion, and intersection scenarios. The model demonstrates strong generalization capabilities by predicting responses in an unseen scenario with fixed parameters.

Key insights

• Active inference provides a theoretically grounded and unified framework capable of explaining complex, multifaceted human behaviors like collision avoidance across various dynamic scenarios.
• Specific cognitive mechanisms, including evidence accumulation, prediction noise, pedal constraints, and norm-conditioned predictions, are critical for reproducing human-like response timing, maneuver choices, and collision outcomes.
• The model demonstrates strong generalizability, accurately predicting human behavior in previously unseen scenarios without requiring parameter re-tuning, supporting its robustness as a comprehensive framework.

Summary by alphaXiv.