Skip to content
RobothecaBeta
Log in

Octo

  • AI model
  • Humanoids and robot learning

Octo is a set of open generalist robot policies: transformer-based diffusion models trained on about 800k robot trajectories, for controlling various robot arms from images and language.

  • Stale

Made by Robotic AI & Learning Lab Berkeley, United States (checked on github.com/octo-models/octo/blob/main/LICENSE (opens in a new tab)).

Source on GitHub: github.com/octo-models/octo (opens in a new tab). Facts read at commit 241fb35 (opens in a new tab).

Licenses

Code
MIT (opens in a new tab)

Read by Robotheca from LICENSE (opens in a new tab) (MIT) at commit 241fb35 (opens in a new tab), on 16 Oct 2026.

Commercial use
Yes

Labels

  • StaleNo commit in 12 months: the newest is from 31 Jul 2024.

What it does

  • Trained on a diverse mix of 800k robot trajectories and built as transformer-based diffusion policies.
  • Accepts multiple RGB camera inputs and can be instructed with language commands or goal images.
  • Can be finetuned to new sensors, action spaces and robot morphologies using a small dataset and modest compute.
  • Pretrained checkpoints Octo-Base (93M params) and Octo-Small (27M params) are available on Hugging Face.
  • Includes example scripts for inference, finetuning, Gym rollouts, and evaluation on a real WidowX robot.
  • Provides code to reproduce pretraining on the Open X-Embodiment data and an Open X dataloader in PyTorch.

Written by Robotheca from the project's README (opens in a new tab) at commit 241fb35.

Getting started

Difficulty
Advanced
Hardware it needs, as the project states it
Inference speed is given for one NVIDIA 4090 GPU (Octo-Base 13 it/sec, Octo-Small 17 it/sec). Pretraining was run on a TPUv4-128 pod (8 hours for Octo-S, 14 hours for Octo-B). Installation is described for GPU (CUDA 11) and TPU setups.

Maintenance

Status
Stale: no commit in 12 months
Newest commit
31 Jul 2024 on main
Latest release
v1.5 (24 May 2024)

Checked on GitHub on 16 Oct 2026.

Links

Robots on Robotheca that use it

No robot on Robotheca is linked to it yet. Links come from a robot's robot.yaml, from its dependency files read as text, or from Robotheca's editors.

Robotheca lists AI models as a reference: nothing is hosted or run here, and every fact links to where it was read.