Flow-ERD: Agent-type Aware Flow Matching with Entropy-Regularized Distillation for Diverse Traffic Simulation

Jul 16, 2026

Upload by Skyler

Realistic and diverse traffic simulation is crucial for autonomous driving development, yet many benchmarks mainly reward realism and leave behavioral diversity underexplored. Flow-ERD presents a multi-agent simulator that jointly optimizes realism and diversity. It uses Agent-Type Aware Flow Matching (AFM) to combine multi-modal expressiveness with type-specific kinematic execution, preserving fine-grained diversity. An Entropy-Regularized Distillation (ERD) stage fine-tunes closed-loop rollouts via entropy-regularized reverse-KL to reduce covariate shift and prevent mode collapse. Evaluation uses standard realism scoring plus a log-free diversity metric and achieves top results on WOSAC.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages8

Reading Time20 min

Introduction

Realism vs. diversity in traffic simulation benchmarks

Generative modeling choices and motion representations

Closed-loop deployment and covariate shift

Proposed method: Flow-ERD (AFM + ERD)

Evaluation on WOSAC

Flow-ERD: Agent-type Aware Flow Matching with Entropy-Regularized Distillation for Diverse Traffic Simulation Seulbin Hwang∗ , Kiyoung Om∗ , Daejung Kim, and Jinhan Lee† arXiv :2607 .06957v 1 [ cs .RO] 8 Jul 2026 Abstract—Realistic and diverse traffic simulation is essential to autonomous driving development. Yet prevailing benchmarks predominantly reward realism, and recent methods have optimized accordingly, leaving diversity underexplored. We introduce Flow-ERD, a multi-agent simulator that pursues realism and diversity jointly. Its backbone, Agent-Type Aware Flow Matching (AFM), couples flow matching’s multi-modal expressiveness with type-specific kinematic execution. It preserves fine-grained diversity while keeping motions consistent with each agent type. A second stage, Entropy-Regularized Distillation (ERD), fine-tunes the closed-loop rollout distribution with an entropy-regularized reverse-KL objective. This mitigates covariate shift while explicitly preventing collapse onto high-density modes. We evaluate Flow-ERD with a log-free diversity metric alongside standard realism scores. Flow-ERD ranks first on the WOSAC test benchmark and dominates the realism–diversity Pareto front among reproducible baselines. Our project page is available here. I. INTRODUCTION Traffic simulation has become core infrastructure for autonomous driving, supporting controlled validation before public-road deployment as well as the development of AV planning policies [1], [2] . For simulation to serve these roles, the surrounding agents, including vehicles, cyclists, and pedestrians, must be realistic, imitating real-world traffic behavior and reacting to one another in a closed loop; and diverse, spanning the multiple plausible futures of a scene to ensure the ego policy’s robustness [3], [4] . These properties must hold jointly, not as alternatives (Fig. 1) . Generative models are well-suited to capturing these properties, learning the distribution of traffic-agent behavior from large-scale data [5] with expressive architectures [6], [7] . Indeed, recent learning-based simulators [8]–[14] have made substantial progress on realism, especially under benchmarks such as the Waymo Open Sim Agents Challenge (WOSAC) [3] . The benchmark’s realism score, however, is measured against a single logged future, and thus cannot distinguish a model that merely fits the logged future from one that captures diverse plausible behaviors. As models are increasingly optimized for this benchmark, diversity has been acknowledged but rarely treated as equally important as realism: it is often left to a sampling hyperparameter [9] or assessed only qualitatively [15], [16] . This gap also manifests in how simulators are designed and trained. By design, backbones trade realism against This work has been submitted to the IEEE for possible publication. Copyright may be transferred without notice, after which this version may no longer be accessible. ∗ Equal contribution;† Corresponding author. All authors are with NAVER LABS Corp., Republic of Korea (e-mail:{[h.sb](h.sb), se99an, [daejung.kim](daejung.kim), [jinhan.lee](jinhan.lee}@naverlabs.com)[}](jinhan.lee}@naverlabs.com)[@naverlabs.com](jinhan.lee}@naverlabs.com)). Fig. 1: Low-diversity rollouts concentrate on a dominant behavior, whereas low-realism rollouts deviate from plausible traffic motion. Flow-ERD targets the desired regime of realistic and diverse closed-loop rollouts. diversity. Next-token-prediction based methods [8]–[10],[17] draw each action from a predefined discrete vocabulary derived from logged data. Because they encode patterns already present in the data, they provide an inductive bias toward realistic, type-compatible motion. However, the fixed vocabulary collapses fine-grained motion onto a coarse set of tokens, which inherently bounds the attainable diversity [10] . Continuous representations [11], [12], [18], especially diffusion models, remove this bottleneck and are

FAQs

What problem does Flow-ERD target in traffic simulation benchmarks?

Flow-ERD targets the mismatch where common benchmarks emphasize realism using logged futures, while diversity is rarely optimized or measured as equally important.

How does Flow-ERD maintain both diversity and realism during generation and rollout?

Flow-ERD uses Agent-Type Aware Flow Matching (AFM) to generate multi-modal actions and executes them through agent-type-specific kinematic transitions, then uses a second-stage Entropy-Regularized Distillation (ERD) to adjust the closed-loop rollout distribution.

How does ERD address covariate shift and prevent collapse?

ERD fine-tunes the closed-loop rollout distribution with an entropy-regularized reverse-KL objective, which both mitigates covariate shift and explicitly discourages collapse onto high-density modes.

Jul 16, 2026

Upload by Skyler