octo-models/octo
A transformer-based generalist robotic policy trained on 800k robot trajectories for multimodal robot arm control.
Not currently ranked — collecting fresh signals.
star history
Octo is a transformer-based diffusion policy that serves as a generalist robot policy. It processes multiple RGB camera inputs and can be controlled via language commands or goal images to generate actions for various robot arms. The model uses a modular attention structure in its transformer backbone, enabling effective finetuning to new robot setups with different sensory inputs, action spaces, and morphologies using limited target domain data.
Frequently asked
- What is octo-models/octo?
- A transformer-based generalist robotic policy trained on 800k robot trajectories for multimodal robot arm control.
- Is octo open source?
- Yes — octo-models/octo is open source, released under the MIT license.
- What language is octo written in?
- octo-models/octo is primarily written in Python.
- How popular is octo?
- octo-models/octo has 1.7k stars on GitHub.
- Where can I find octo?
- octo-models/octo is on GitHub at https://github.com/octo-models/octo.