mimic-video: Video-Action Models for Generalizable Robot Control Beyond VLAs
Original research
MODEL PROFILE · World-action research
mimic-video
Decodes robot actions from the latent trajectory generated by a pretrained video model.
Research organisationPai and collaborators
Research approachInverse dynamics
Published year2025
BackbonePretrained video model / flow decoder
Overview
mimic-video pairs a pretrained video model with a flow-matching inverse-dynamics action decoder reading the model’s latent plan. It is a video-action method; the primary paper supplies the evidence.
Source record
Core research identity and broad method were source-checked as of 2026-11-03. Independent benchmarking, commercial availability, weights and licensing have not been assessed.