ReactSim-Bench: Benchmarking Reactive Behavior World Model Simulation in Autonomous Driving

2026-06-11

AI Summary

This work addresses a critical gap in existing autonomous driving simulation benchmarks, which lack evaluation of surrounding agents’ ability to respond realistically to ego-vehicle behaviors that deviate from logged trajectories. To this end, the authors propose ReactSim-Bench, the first benchmark specifically designed to assess the reactive capabilities of world models. By decoupling ego-vehicle and surrounding-agent control, the framework uses planner-generated off-log ego behaviors as input and constructs 2,636 validated test cases spanning three scenario categories through rule-based filtering and human verification. It further introduces multidimensional safety-compliance metrics, including collision rates, map adherence, and kinematic feasibility. Experiments demonstrate that ReactSim-Bench effectively exposes limitations in the reaction plausibility of current world models based on Transformers, diffusion architectures, and next-token prediction paradigms, and reveal that replanning frequency significantly influences simulation fidelity.

Abstract

Reactive capability is a key property of data-driven behavior world model simulators for autonomous driving simulation systems. With this capability, simulated world agents can respond feasibly to autonomous vehicle (AV) behaviors that differ from the log. However, existing behavior simulation benchmarks do not directly measure reactive capability. They often let the simulator jointly control the AV and surrounding agents and evaluate realism through log similarity or open-loop prediction metrics. In this work, we introduce ReactSim-Bench for evaluating the reactive capability of behavior world model simulation in autonomous driving. We decouple the control of agents and the AV, using AV behaviors that differ from the log and require agents to respond as independent AV inputs. To obtain these AV behaviors, we construct a pipeline that uses an AV planner model to generate candidate behaviors and filters the data using rules and manual verification. Collision metrics, map-based metrics, and kinematic feasibility metrics are used to evaluate the safety and rule compliance of reactive responses. We construct 2,636 test scenarios with three categories and conduct a systematic evaluation of state-of-the-art models across multiple architectures, including Transformer-based, diffusion-based, and next-token-prediction-based models. We further analyze how replan frequency affects performance and provide insights for future studies.

Problem

• reactive capability
• behavior world model
• autonomous driving simulation
• simulation benchmark
• agent response

Innovation

• reactive simulation
• behavior world model
• autonomous driving benchmark
• agent-AV decoupling
• reactive capability evaluation