UniPi: Learning universal policies via text-guided video generation
Original research
MODEL PROFILE · World-action research
UniPi
An early world-action recipe: imagine a future video, then infer the actions to realise it.
Research organisationGoogle Research
Research approachInverse dynamics
Published year2023
BackboneVideo diffusion
Overview
UniPi was an early demonstration of video-based robotic planning. Its approach predicts visual futures before recovering robot actions, and is described in NVIDIA’s historical comparison of WAM formulations.
Source record
- UniPi: Learning universal policies via text-guided video generation
- NVIDIA WAM technical synthesis (Moritz Reuss, 15 Jun 2026)
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.