Decision Making for Autonomous Vehicles at Unsignalized Intersection in Presence of Malicious Vehicles

Pruekprasert, Sasinee; Zhang, Xiao–Yi; Dubut, Jérémy; Huang, Chao; Kishida, Masako · 2019 · OpenAlex-citations

DOI: 10.48550/arxiv.1904.10158

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 decision-making for autonomous vehicles (AVs) at unsignalized intersections, specifically in scenarios involving "malicious" vehicles that violate right-of-way rules or exhibit irrational behavior. Motivated by the high frequency of collisions at unsignalized intersections and the need for AVs to safely share roads with human drivers who may not adhere to traffic laws, the authors propose a game-theoretic framework. The study aims to enable an "angelic" AV (one that follows the law) to navigate safely and efficiently despite the presence of aggressive or unpredictable agents. The methodology employs a finite perfect-information game where vehicles determine their acceleration inputs by computing Nash equilibria. Unlike previous approaches relying on estimated "aggressiveness" values, this method bases priority orders on formal right-of-way rules (specifically Japanese regulations) and the vehicle's behavioral type. The authors classify vehicles into four types: angelic (law-abiding), intermediate (selfish but adaptable), demonic (selfish and rigid), and irrational (random actions). Each vehicle maintains a priority order initialized based on its type and updated dynamically. Angelic and intermediate vehicles update their priority beliefs when predictions of other vehicles' actions deviate from observations, allowing them to fit their models to the actual behavior of malicious agents. The cost function minimizes safety risks and velocity violations, while a specific mechanism introduces randomness to resolve potential deadlocks where all vehicles wait for others to proceed. The authors validated the approach through numerical simulations in MATLAB, running 2,000 trials per scenario with four vehicles approaching the intersection. They tested four cases: all angelic vehicles, three angelic and one demonic, all intermediate, and three intermediate with one irrational vehicle. Results indicated that the angelic AV successfully avoided collisions in all scenarios involving compliant or predictable malicious agents (Cases 1, 2, 3, and their variants). However, in the presence of irrational vehicles (Case 4), the collision rate rose to 0.4%–1.1%, and congestion increased significantly, with average total time steps rising from approximately 54 to 92 steps. The system demonstrated robustness against demonic vehicles, maintaining low collision rates and moderate congestion, but struggled with the unpredictability of irrational agents. The significance of this work lies in its formalization of priority-based decision-making that adapts to rule-breaking behaviors without requiring vehicle-to-vehicle communication. By integrating right-of-way rules with game-theoretic equilibrium computation, the approach provides a scalable solution for complex intersection scenarios. The findings highlight that while game-theoretic methods can handle selfish drivers effectively, they face limitations when dealing with truly irrational behaviors, suggesting a need for further refinement in handling extreme unpredictability. This contributes to the development of safer autonomous systems capable of operating in mixed-traffic environments.

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