Generative Image as Action Models
Original research
MODEL PROFILE · Adjacent embodied policy
GENIMA
Draws action targets into images and maps those targets to robot joint positions.
Research organisationDyson Robot Learning Lab
Research approachImage-action generation
Published year2024
BackboneStable Diffusion
Overview
GENIMA fine-tunes Stable Diffusion to generate images containing visual joint-action targets. A separate controller transforms these into robot joint commands. This is an image-action policy approach, rather than a conventional future-video WAM.
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.