Unleashing Large-Scale Video Generative Pre-training for Visual Robot Manipulation
Original research
MODEL PROFILE · World-action research
GR-1
Pretrains a video-predictive transformer before adapting it to language-conditioned robot actions.
Research organisationByteDance Research
Research approachJoint prediction
Published year2023
BackboneGPT-style video pretraining
Overview
GR-1 jointly predicts robot actions and future images using a GPT-style model pretrained on video and subsequently adapted for manipulation. Its source paper describes the architecture and evaluation; reported outcomes have not been independently replicated here.
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.