UniDBO: A Unified Dual-Branch One-Step Denoising Framework
for Autonomous Driving Scenario Generation

Traffic scenario generation is essential for training, testing, and safety validation of autonomous vehicles (AVs), especially when real-world data coverage is limited. Existing methods mostly face a three-way trade-off among open-loop trajectory fitting, closed-loop simulation robustness, and inference efficiency. Autoregressive methods are typically efficient in open-loop fitting but susceptible to error accumulation and policy-induced state-distribution shift in long-horizon closed-loop rollouts. Diffusion-based methods provide strong multi-modal behavior modeling and competitive generation quality, but iterative denoising substantially increases inference latency and limits simulation throughput. Moreover, repeated refinement under a fixed scene condition is not fully aligned with the continuous feedback and frequent replanning required by closed-loop simulation. To address this challenge, we propose UniDBO, a Unified Dual-Branch One-step denoising framework for autonomous driving scenario generation, which uses a shared scene encoder and two jointly optimized complementary branches with distinct roles—a continuous-scale branch (CS) for open-loop trajectory fitting and a discrete high-noise branch (DHN) for robust closed-loop simulation. Through joint training, UniDBO coordinates open-loop fitting and closed-loop robustness while preserving one-step inference efficiency. We evaluate UniDBO with different benchmarks, covering in-domain open-loop and closed-loop evaluation, zero-shot cross-domain generalization, and system-level closed-loop AV testing in unseen scenarios. Results show that UniDBO improves the open-loop/closed-loop balance while maintaining high inference efficiency, and achieves competitive zero-shot generalization on the evaluated benchmarks.