EEmbodied AI Hub
ModelFully open

RoboFlamingo

Low-cost imitation approach built on open VLMs for language-conditioned manipulation.

Official / sourcePaperBack to Model
VLAOpen sourceImitation learning

Overview

RoboFlamingo adapts Flamingo-style vision-language models to robot manipulation, emphasizing open-ended language-conditioned control with relatively accessible open components for research. It is part of the broader “VLM → robot” transition that led to modern VLAs.

Who it is for

Researchers studying VLM-based manipulation policies.

Key highlights

  • Flamingo-style robotics adaptation
  • Language-conditioned manipulation
  • Research reference implementation

When to use

  • VLM-policy ablations
  • Historical baselines in VLA papers

When not to use

  • You want the most maintained production stack in 2026

Getting started

  1. 1Follow repo install
  2. 2Run language-conditioned eval
  3. 3Compare to OpenVLA on shared tasks

Papers & reading

Caveats & pitfalls

  • May lag newer training tooling.

Content reviewed 2026-07-23