Compact Latent Coordination for Autonomous Vehicles at Unsignalized Intersections

2026-07-23

AI Summary

This work addresses the challenges of multi-agent coordination among autonomous vehicles at unsignalized intersections, including the large combinatorial action space, reliance on privileged information, and inflexible agent design. To overcome these issues, the authors propose MAPS, a hierarchical deep reinforcement learning framework that decouples strategic intent from tactical execution. A centralized Master generates a compact, continuous global coordination embedding—termed a proto-plan—while decentralized Workers execute control actions based on local observations. The proto-plan mechanism enables separation of policy and execution, facilitates modular optimization, and confers strong zero-shot generalization when the number of agents at deployment differs from training. Experiments demonstrate collision-free navigation across 72 intersection configurations with significantly reduced average travel time; notably, a model trained with three vehicles achieves a 94% success rate when zero-shot transferred to a five-vehicle scenario.

Abstract

Coordinating autonomous vehicles at unsignalized intersections remains a critical challenge for multi-agent reinforcement learning (MARL) systems, which typically struggle with combinatorial action spaces, reliance on privileged information, or rigid agent designs. We propose Master-Agent Proto-plan System (MAPS), a hierarchical deep reinforcement learning (DRL) architecture in which a centralized Master agent generates a compact, continuous embedding, denoted as proto-plan, that encodes a global coordination strategy. Decentralized Worker agents integrate this embedding with local observations to execute vehicle-specific control, decoupling strategic intent from tactical execution and enabling independent optimization of each module. As a proof-of-concept evaluation of this coordination mechanism, we test MAPS across 72 intersection configurations in HighwayEnv. MAPS achieves collision-free navigation while significantly reducing average travel time, outperforming state-of-the-art baselines. The learned proto-plans further exhibit robust generalization: a system trained with three agents achieves a 94% success rate when deployed zero-shot to five-agent scenarios, confirming that proto-plan-based hierarchical learning provides a promising framework for multi-vehicle coordination.

Problem

• autonomous vehicles
• unsignalized intersections
• multi-agent reinforcement learning
• coordination
• combinatorial action spaces

Innovation

• proto-plan
• hierarchical reinforcement learning
• multi-agent coordination
• autonomous vehicles
• unsignalized intersections