DISK: Dynamic Inference SKipping for World Models

TL;DR

DISK is a training-free adaptive inference method for autoregressive world models that significantly speeds up video and trajectory diffusion while maintaining accuracy and quality, enabling efficient long-horizon predictions.

Contribution

DISK introduces a novel, training-free inference approach that coordinates dual diffusion transformers with cross-modal skip decisions for improved efficiency in world models.

Findings

2x speedup on trajectory diffusion

1.6x speedup on video diffusion

Maintains accuracy and visual quality

Abstract

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.