Single Diffusion Model for Autonomous Driving Planning and Scenario Generation
September 7, 2026
A Single-Stream Dual-Stream (SSDS) diffusion-transformer decoder enables a single pretrained model to function as both an ego motion planner and a controllable scenario generator. The method uses Decoupled Annealing Posterior Sampling with Energy (DAPSE) to inject energy functions without auxiliary networks, improving closed-loop nuPlan performance.
HOW THIS AFFECTS YOU
●
builderYou can use one model to both drive agents and stress-test them in simulation.
●
researcherYou can utilize the SSDS architecture to fuse scene context via joint attention rather than late cross-attention.