DIDO: One-Step Denoising for World Action Models
Loading video
Loading videoDIDO distills a four-step video world model into one denoising step, preserving gripper-object dynamics for manipulation while cutting action latency from 562 ms to 384 ms.
Category: arm
Author: @stepjamUK
Date: 2026-09-15T00:00:00
Duration: 68.091s
Reference: https://arxiv.org/abs/2609.15570
