Adnan KhanAdnanProduct & AI
San Francisco Bay Area
Get in touch

Source note2 min read

Compact Latent Coordination for Autonomous Vehicles at Unsignalized Intersections

Source

arXiv

arxiv.org/abs/2607.21488v1

Read the source

Source

Plain-Language Brief

A new approach called Master-Agent Proto-plan System (MAPS) is proposed to improve the coordination of autonomous vehicles at unsignalized intersections. MAPS uses a hierarchical deep reinforcement learning architecture, where a central agent generates a global strategy and decentralized agents execute vehicle-specific control. The study shows that MAPS achieves collision-free navigation and reduces travel time compared to state-of-the-art baselines in 72 intersection configurations. The learned strategies also generalize well, performing effectively in scenarios with more vehicles.

What the Document Says

  • The text discusses a new coordination mechanism for autonomous vehicles at unsignalized intersections called Master-Agent Proto-plan System (MAPS).
  • MAPS is a hierarchical deep reinforcement learning architecture with a centralized Master agent and decentralized Worker agents.
  • The Master agent generates a global strategy, represented as a compact, continuous embedding, or proto-plan.
  • Worker agents integrate the proto-plan with local observations to execute vehicle-specific control.
  • This decouples strategic intent from tactical execution and allows independent optimization of each module.
  • The study tests MAPS across 72 intersection configurations in HighwayEnv.
  • MAPS achieves collision-free navigation and significantly reduces average travel time compared to state-of-the-art baselines.
  • The learned proto-plans show robust generalization, with a system trained with three agents achieving a 94% success rate in five-agent scenarios.
  • The text does not provide specific details about the state-of-the-art baselines used for comparison.

Key Concepts Mentioned

  • autonomous vehicles
  • unsignalized intersections
  • multi-agent reinforcement learning (MARL)
  • hierarchical deep reinforcement learning (DRL)
  • Master agent
  • Worker agents
  • proto-plan
  • collision-free navigation
  • travel time
  • generalization

Claims Made by the Document

  • MAPS improves coordination of autonomous vehicles at unsignalized intersections.
  • MAPS reduces average travel time compared to state-of-the-art baselines.
  • The learned proto-plans in MAPS generalize well.

Evidence Used

  • The text mentions that MAPS achieves collision-free navigation and significantly reduces average travel time compared to state-of-the-art baselines in 72 intersection configurations.
  • The text states that a system trained with three agents achieves a 94% success rate when deployed zero-shot to five-agent scenarios, demonstrating the learned proto-plans' generalization.

Methods / Approach

UNKNOWN

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

UNKNOWN

Open Questions

UNKNOWN

Notes for Future Q&A

May include interpretation. Labeled where it does.

UNKNOWN

Raw Abstract / Excerpt

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.