Search papers, labs, and topics across Lattice.
The paper introduces DISK, a training-free adaptive inference method for autoregressive world models that coordinates video and ego-trajectory diffusion transformers using dual-branch controllers with cross-modal skip decisions. DISK extends higher-order latent-difference skip testing to autoregressive chains and propagates controller statistics for long-horizon stability, addressing the computational expense of long-horizon video and trajectory prediction. Experiments on NuPlan and NuScenes datasets demonstrate that DISK achieves 2x speedup on trajectory diffusion and 1.6x speedup on video diffusion while maintaining performance metrics like L2 planning error, FID/FVD, and NAVSIM PDMS scores.
Double the speed of trajectory prediction and 1.6x faster video diffusion in driving world models, without sacrificing accuracy, thanks to a training-free adaptive skipping method.
We present DISK, a training-free adaptive inference method for autoregressive world models. DISK coordinates two coupled diffusion transformers for video and ego-trajectory via dual-branch controllers with cross-modal skip decisions, preserving motion-appearance consistency without retraining. We extend higher-order latent-difference skip testing to the autoregressive chain-of-forward regime and propagate controller statistics through rollout loops for long-horizon stability. When integrated into closed-loop driving rollouts on 1500 NuPlan and NuScenes samples using an NVIDIA L40S GPU, DISK achieves 2x speedup on trajectory diffusion and 1.6x speedup on video diffusion while maintaining L2 planning error, visual quality (FID/FVD), and NAVSIM PDMS scores, demonstrating practical long-horizon video-and-trajectory prediction at substantially reduced cost.
