Generalist robot policy Transformer with multi-modal inputs and multi-embodiment finetuning.
Octo is a common generalist policy baseline in open robot learning. Useful when you want a research comparison point alongside OpenVLA-style models. Finetune and eval still take care; not the lightest onboarding path. Intermediate.5 — https://github.com/octo-models/octo/releases/tag/v1.5.5 — https://github.com/octo-models/octo/releases/tag/v1.5.5 — https://github.com/octo-models/octo/releases/tag/v1.5. Latest GitHub release: v1.5 — https://github.com/octo-models/octo/releases/tag/v1.5.
Octo is a generalist robot policy transformer trained across diverse robot datasets (OXE-scale). It is a standard open baseline for cross-embodiment finetuning and language-conditioned control experiments.
Read the model card for observation spaces and finetune scripts. For many labs it is the “open generalist” checkpoint before proprietary VLAs.
Learning Path: core VLA-stage generalist baseline.
Labs needing lighter generalist policies; multi-robot finetune studies.
Physical Intelligence generalist VLA (~3B) with flow-matching actions; π0 weights are partially open — strong dexterous-manipulation reference in the π0 family.
Content reviewed 2026-07-28
