arXiv:2512.07661v2 Announce Type: replace 
Abstract: Realistic and diverse multi-agent driving scenes are crucial for evaluating autonomous vehicles, but safety-critical events which are essential for this task are rare and underrepresented in driving datasets. Data-driven scene generation offers a low-cost alternative by synthesizing complex traffic behaviors from existing driving logs. However, existing models often lack controllability or yield samples that violate physical or social constraints, limiting their usability. We present OMEGA, an optimization-guided, training-free framework that enforces structural consistency and interaction awareness during diffusion-based sampling from a scene generation model. OMEGA re-anchors each reverse diffusion step via constrained optimization, steering the generation towards physically plausible and behaviorally coherent trajectories. Building on this framework, we formulate ego-attacker interactions as a game-theoretic optimization in the distribution space, approximating Nash equilibria to generate realistic, safety-critical adversarial scenarios. Experiments on nuPlan and Waymo show that OMEGA improves generation realism, consistency, and controllability, increasing the ratio of physically and behaviorally valid scenes from 32.35% to 72.27% for free exploration capabilities, and from 11% to 80% for controllability-focused generation. Our approach can also generate $5\times$ more near-collision frames with a time-to-collision under three seconds while maintaining the overall scene realism.

تم تقديم إطار جديد يسمى OMEGA لتوليد مشاهد قيادة متعددة الوكلاء بشكل واقعي، مما يعالج ندرة الأحداث الحرجة للسلامة في مجموعات البيانات الحالية. تعزز هذه الطريقة الموجهة بواسطة التحسين والتي لا تتطلب تدريبًا من التحكم والموثوقية الفيزيائية للسلوكيات المرورية المولدة، والتي تعتبر ضرورية لتقييم المركبات الذاتية القيادة.

Se ha presentado un nuevo marco llamado OMEGA para generar escenas de conducción multiagente realistas, abordando la escasez de eventos críticos para la seguridad en los conjuntos de datos existentes. Este enfoque guiado por optimización y sin necesidad de entrenamiento mejora la controlabilidad y la plausibilidad física de los comportamientos de tráfico generados, que son esenciales para evaluar vehículos autónomos.

Un nouveau cadre nommé OMEGA a été introduit pour générer des scènes de conduite multi-agents réalistes, abordant la rareté des événements critiques pour la sécurité dans les ensembles de données existants. Cette approche guidée par l'optimisation et sans formation améliore la contrôlabilité et la plausibilité physique des comportements de circulation générés, essentiels pour évaluer les véhicules autonomes.

A new framework named OMEGA has been introduced for generating realistic multi-agent driving scenes, addressing the scarcity of safety-critical events in existing datasets. This optimization-guided, training-free approach enhances the controllability and physical plausibility of generated traffic behaviors, which are essential for evaluating autonomous vehicles.

Optimization-Guided Diffusion for Interactive Scene Generation

Was this article worth reading? Share it

LucidQuery AI

GPTHumanizer

Scop.ai

DreamGen

Synthesia

AiReelGenerator.com

Ready to build your own newsroom?