Compact Latent Coordination for Autonomous Vehicles at Unsignalized Intersections

2026-07-23 · Source: Machine Learning · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Autonomous Vehicles & Smart Transportation · Depth: Expert, quick

Summary

The Master-Agent Proto-plan System (MAPS) is a hierarchical deep reinforcement learning (DRL) architecture designed to coordinate autonomous vehicles at unsignalized intersections, addressing challenges like combinatorial action spaces and reliance on privileged information in multi-agent reinforcement learning (MARL). MAPS features a centralized Master agent that generates a compact, continuous "proto-plan" embedding, encoding a global coordination strategy. Decentralized Worker agents then integrate this proto-plan with local observations for vehicle-specific control, effectively decoupling strategic intent from tactical execution. Evaluated across 72 intersection configurations in HighwayEnv, MAPS achieved collision-free navigation and significantly reduced average travel time, outperforming existing baselines. Furthermore, a system trained with three agents demonstrated robust generalization, achieving a 94% success rate when deployed zero-shot to five-agent scenarios, confirming its potential for multi-vehicle coordination.

Key takeaway

For Machine Learning Engineers developing multi-agent systems for autonomous driving, consider adopting hierarchical DRL architectures like MAPS. This approach, which separates global strategic intent via a "proto-plan" from local tactical execution, offers superior generalization and performance for complex coordination tasks at unsignalized intersections. You can achieve collision-free navigation and significantly reduce travel times, even when scaling agent numbers zero-shot.

Key insights

MAPS uses hierarchical DRL with a proto-plan embedding for robust, generalizable coordination of autonomous vehicles at unsignalized intersections.

Principles

• Decouple strategic intent from tactical execution.
• Centralized strategy, decentralized execution.
• Compact embeddings enable robust generalization.

Method

A Master agent generates a continuous proto-plan embedding for global coordination. Worker agents integrate this proto-plan with local observations to execute vehicle-specific control, optimizing modules independently.

In practice

• Coordinate autonomous vehicles at unsignalized intersections.
• Scale multi-agent systems zero-shot (e.g., 3 to 5 agents).

Topics

• Autonomous Vehicles
• Multi-Agent Reinforcement Learning
• Hierarchical DRL
• Vehicle Coordination
• Unsignalized Intersections
• Proto-plan

Best for: Computer Vision Engineer, Research Scientist, AI Scientist, Machine Learning Engineer, Robotics Engineer

Related on AIssential

• Beyond Conservative Automated Driving in Multi-Agent Scenarios via Coupled Model Predictive Control and Deep Reinforcement Learning — Coupling RL for speed guidance with MPC for trajectory optimization improves autonomous driving safety and efficiency.
• PLAN-S: Bridging Planning with Latent Style Dynamics for Autonomous Driving World Models — PLAN-S improves autonomous driving LWMs by using style-conditioned semantic cost maps for explicit risk and style modulation in planning.
• Towards End to End Motion Planning and Execution for Autonomous Underwater Vehicles Using Reinforcement Learning — An end-to-end hierarchical DRL system enables AUV navigation from raw sensor data to thruster commands.
• A Generative Model for Closed-Loop Microsimulation of Signalized Intersections — Actor-centric generative model Enactor enables stable, realistic closed-loop microsimulation of signalized intersections by learning from its own predictions.
• Shift & Drift: A Zero-Shot Benchmark for Generalizable and Robust Autonomous Driving Motion Planning — The benchmark reveals that current AD planners, especially IL-based, lack generalization and robustness to semantic and state-distribution shifts.
• World Models as Adversaries: Multi-Agent Self-Play Fine-Tuning for Robust Motion Planning — AWM uses multi-agent self-play and a decoupled min-max game to fine-tune robust motion planners for autonomous vehicles in rare, critical scenarios.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.