Long-term Traffic Simulation with Interleaved Autoregressive Motion and Scenario Generation

2025-06-20

AI Summary

Existing traffic simulators primarily focus on closed-loop motion prediction for initially present vehicles, making them inadequate for long-horizon (e.g., 30-second) dynamic scenario simulation required by autonomous driving—vehicle entry and exit over time cause conventional methods to fail. This paper introduces InfGen, the first interleaved autoregressive modeling framework that unifies closed-loop motion prediction and scene generation with automatic mode switching. Its core innovations include: (1) joint motion-scene tokenization based on next-token prediction; (2) an alternating masking mechanism to seamlessly interleave vehicle-level motion forecasting and scene-level generation; and (3) end-to-end closed-loop training. Experiments show InfGen achieves state-of-the-art performance at 9-second horizons and significantly outperforms all baselines at 30 seconds. Notably, it is the first method enabling point-to-point, trip-level simulation. Code and models are publicly released.

Technology Category

• Search and Optimization: Sampling/Simulation-based Search
• Natural Language Processing: Generation
• Planning, Routing, and Scheduling: Planning with Language Models

Application Category

• User Modeling, Personalization and Recommendation: User modeling and simulation for interactive and conversational systems
• Web Mining and Content Analysis: Web data generation and simulation
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

An ideal traffic simulator replicates the realistic long-term point-to-point trip that a self-driving system experiences during deployment. Prior models and benchmarks focus on closed-loop motion simulation for initial agents in a scene. This is problematic for long-term simulation. Agents enter and exit the scene as the ego vehicle enters new regions. We propose InfGen, a unified next-token prediction model that performs interleaved closed-loop motion simulation and scene generation. InfGen automatically switches between closed-loop motion simulation and scene generation mode. It enables stable long-term rollout simulation. InfGen performs at the state-of-the-art in short-term (9s) traffic simulation, and significantly outperforms all other methods in long-term (30s) simulation. The code and model of InfGen will be released at https://orangesodahub.github.io/InfGen

Problem

• Simulating long-term realistic traffic for self-driving systems
• Handling agent entry and exit in dynamic traffic scenarios
• Unifying motion simulation and scene generation for stability

Innovation

• Unified next-token prediction model for traffic simulation
• Interleaved closed-loop motion and scene generation
• Automatic switching between simulation and generation modes