A Generative Model for Closed-Loop Microsimulation of Signalized Intersections

2026-06-22 · Source: Takara TLDR - Daily AI Papers · Field: Transportation & Mobility — Autonomous Vehicles & Smart Transportation, Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, medium

Summary

Enactor is an actor-centric generative model designed for closed-loop microsimulation of signalized intersections, addressing limitations of traditional traffic models and short-horizon trajectory predictors. It focuses on vehicle behavior, incorporating pedestrians as contextual influences, and encodes dynamic actors and lane polylines using polar coordinates. The model employs a transformer architecture with distinct spatial and temporal attention blocks to predict distributions for each actor's next-step motion ($s$, $α$). Trained with a closed-loop curriculum, Enactor was evaluated in a 4000-second simulation-in-the-loop test across two intersection geometries. It accurately recovers SUMO data generator's speed and travel-time distributions, achieving KL divergence an order of magnitude lower than a transformer baseline for travel time, and approximately $5\times$ lower for speed at Site 1. Furthermore, Enactor significantly reduces red-light violations by over an order of magnitude and outperforms a constant-velocity baseline on multi-horizon predictive tasks using real-world field data.

Key takeaway

For Machine Learning Engineers developing autonomous driving simulation environments, Enactor demonstrates a robust approach to closed-loop traffic microsimulation. You should consider adopting actor-centric generative models with closed-loop curriculum training to achieve higher fidelity in vehicle interaction and reduce simulation instability. This method significantly improves realism in speed and travel-time distributions while drastically cutting red-light violations, offering a more reliable platform for policy evaluation.

Key insights

Actor-centric generative model Enactor enables stable, realistic closed-loop microsimulation of signalized intersections by learning from its own predictions.

Principles

• Closed-loop curriculum training improves model stability and realism.
• Actor-centric modeling with contextual elements enhances interaction capture.
• Polar coordinate encoding effectively represents dynamic actors and lane polylines.

Method

Encode dynamic actors and lane polylines in polar coordinates. Employ a transformer with separate spatial and temporal attention blocks to predict actor motion. Train using a closed-loop curriculum, exposing the model to its own predictions.

In practice

• Incorporate leader rear-bumper features for enhanced intersection safety.
• Evaluate simulation models against continuously refreshing actor sets.

Topics

• Generative Models
• Traffic Simulation
• Closed-Loop Systems
• Signalized Intersections
• Autonomous Vehicles
• Transformer Architecture

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

Related on AIssential

• CLEAR: Cognition and Latent Evaluation for Adaptive Routing in End-to-End Autonomous Driving — CLEAR achieves real-time, multi-modal autonomous driving by replacing iterative denoising with single-step latent space drift and semantic reasoning.
• Improving Crash Frequency Prediction from Simulated Traffic Conflicts Using Machine Learning Based Microsimulation — ML-based traffic microsimulation improves crash frequency prediction from simulated conflicts without location-specific calibration.
• PersonaDrive: Human-Style Retrieval-Augmented VLA Agents for Closed-Loop Driving Simulation — PersonaDrive enables style-diverse VLA driving agents by conditioning them on retrieved human demonstrations, eliminating per-style retraining.
• Beyond Conservative Automated Driving in Multi-Agent Scenarios via Coupled Model Predictive Control and Deep Reinforcement Learning — Integrating MPC and RL improves automated driving safety, efficiency, and generalization in multi-agent scenarios.
• DriveCtrl: Conditioned Sim-to-Real Driving Video Generation — DriveCtrl bridges the sim-to-real gap for autonomous driving by generating realistic, temporally consistent, and annotation-preserving videos.
• Scaling Self-Play for End-to-End Driving — Self-play DAgger training in a high-throughput simulator enables scalable, robust end-to-end autonomous driving from pixels.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Takara TLDR - Daily AI Papers.