Back to search

University of Glasgow

Safety-critical decision making and coordination for autonomous vehicles in mixed traffic

Abstract

dc:description.abstract

Coordinating multiple autonomous vehicles at unsignalized intersections remains a fundamental challenge in multi-agent systems. The exponential growth of joint action spaces, coordination ambiguity under symmetric configurations, and stringent real-time constraints render centralized approaches intractable. This thesis introduces a framework that reformulates Level-k cognitive hierarchy for safety-critical coordination, integrating with Monte Carlo Tree Search (MCTS) to achieve scalable planning with emergent safety properties. The central contribution redefines Level-0 as a universal safety initialization generating conservative baseline trajectories, rather than modeling naive random behaviors. This transforms Level-k reasoning from a descriptive cognitive model into a constructive planning framework where safety emerges structurally through cascading conservative margins: Level 1 agents inherit Level-0 safety anchors while optimizing efficiency, and Level-2 agents amplify these margins by anticipating Level-1 strategic responses. The framework decomposes multi-agent coordination via dual-filtered interaction graphs combining spatial conflict detection with strategic reasoning, reducing computational complexity from exponential to linear in agent count. The MCTS integration enables efficient exploration of the action space through selective sampling guided by Upper Confidence Bounds. Safety-aware pruning eliminates approximately 70% of unsafe actions during tree expansion, reducing the effective branching factor from 15 to approximately 4–5 actions per node. Trajectory caching exploits the deterministic nature of Level-k rollouts to achieve 35% cache hit rates, avoiding redundant computation. For mixed traffic scenarios involving human-driven vehicles, the framework incorporates style-aware behavior prediction based on the Intelligent Driver Model, time-varying uncertainty quantification, and adaptive safety thresholds that respond to interaction-specific risks. The complete framework reduces computational complexity over 20 orders of magnitude compared to joint optimization, enabling sub-100-millisecond planning cycles suitable for real-time deployment. Extensive experimental validation across challenging scenarios demonstrates the framework’s effectiveness. In symmetric eight-agent intersection coordination where baseline methods exhibit 15–35% collision rates, the proposed approach achieves zero collisions with 95–98% arrival rates. Mixed traffic experiments at 50% autonomous vehicle penetration maintain collision rates below 2% despite diverse human driving behaviors, with consistent performance across penetration rates from 20% to 100%. The framework’s interpretability through explicit reasoning traces, modularity enabling component-wise validation, and fully decentralized architecture requiring no inter-vehicle communication provide practical advantages for real-world deployment. Beyond autonomous driving, the theoretical contributions—reconstructed Level-k reasoning with emergent safety and dual-filtered interaction decomposition—offer broader insights applicable to multi-agent coordination challenges across robotics, game theory, and artificial intelligence.

Degree

thesis:*
Level dc:type.qualificationlevel
PhD
Grantor dc:publisher.institution
University of Glasgow
Year dc:date.issued
2026

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Lin, Zhihao

Subjects

dc:subject × 2

Rights

Language dc:language
en

Identifiers

dc:identifier.*

Chain of custody

source
Harvested from
University of Glasgow
Base URL
theses.gla.ac.uk/cgi/oai2
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Lin, Zhihao. Safety-critical decision making and coordination for autonomous vehicles in mixed traffic. PhD thesis, University of Glasgow, 2026. https://doi.org/10.5525/gla.thesis.86066