Bridging Local Observation and Global Simulation in Closed-Loop Traffic Modeling

2026-06-30

AI Summary

This work addresses the challenge that autoregressive traffic simulators, when trained on egocentric logs, often exhibit anomalous behaviors in globally observable closed-loop environments due to mismatches between local observations and global context. To mitigate this issue, the authors propose CRAFT, a novel framework that introduces test-time preference alignment into traffic simulation. CRAFT employs a self-supervised failure discovery mechanism to generate diverse rollouts that expose context-induced errors, and leverages human-aligned driving priors to construct preference supervision. This enables training a Context Preference Evaluator (CPE) that reweights autoregressive action decoding during inference. Notably, CRAFT enhances behavioral plausibility without requiring retraining of the base model, achieving a 31.2% reduction in collision rates and a 33.2% decrease in traffic violations in closed-loop simulation.

Abstract

A local-to-global context mismatch arises when autoregressive traffic simulators trained on ego-centric driving logs are deployed in globally observable closed-loop environments. In such logs, the ego vehicle has rich local observations, while surrounding agents are only partially observed due to perception limits and occlusions. As a result, simulators may learn incomplete context--action mappings that remain hidden in log-based training but emerge during closed-loop rollouts, leading to unrealistic behaviors such as abnormal stops, unsafe interactions, and rule violations. We propose CRAFT, a Contextual pReference Alignment Framework for Traffic Simulation, to mitigate this mismatch via self-supervised failure discovery and preference-guided test-time alignment. CRAFT treats the base simulator as a globally observable sandbox, generating diverse what-if rollouts from logged initial states to expose context-induced failures. These failures are grounded with human-aligned driving priors and converted into preference supervision for training a Contextual Preference Evaluator (CPE). At inference time, CPE acts as a plug-in alignment module that scores candidate actions under complete scene context and reweights autoregressive decoding toward globally coherent behaviors. CRAFT mitigates this local-to-global contextual bias, reducing collisions by 31.2\% and traffic violations by 33.2\% without retraining the base simulator.

Problem

• context mismatch
• closed-loop traffic simulation
• autoregressive simulators
• ego-centric logs
• unrealistic behaviors

Innovation

• closed-loop traffic simulation
• context mismatch
• preference alignment
• self-supervised failure discovery
• autoregressive decoding