Octo
Generalist robot policy trained on 800K trajectories
Octo was trained by watching 800,000 robot movements across dozens of different robots. It learned patterns that apply broadly — like how to approach an object, grip it, and release it. You can then teach it your specific task with just a few examples.
Relatively accessible — runs on a mid-range GPU and has solid documentation. You'll need some Python and ML knowledge, but it's one of the easier research models to get running.
Transformer-based diffusion policy trained on 800K robot trajectories from Open X-Embodiment. Designed as a generalist policy that can be fine-tuned for new tasks with minimal data.
Open X-Embodiment (800K trajectories, multiple robot types)
Standard baseline for manipulation papers
Bin-picking with minimal fine-tuning data
Teaching robots from human demonstrations
Requires some ML and Python experience
RELATED DISCUSSIONS
Community →Seeded and planned prompts — not live forum activity yet.
- Community signalIntermediate / AdvancedHow should EAR track open robotics models?
Meta-discussion on ontology, benchmarks, and graph links for open-weight VLAs and manipulation models.
- Community signalIntermediate / AdvancedOpenVLA in the field — what breaks first?
Open Models room prompt linked to openvla-7b — inference, fine-tuning, and sim-to-real gaps. Seeded prompt only.
Related Models
VIEW ALL →LeRobot ACT
Action Chunking Transformer for bimanual manipulation
Diffusion Policy
Visuomotor policy learning via conditional diffusion
OpenVLA
State-of-the-art open-source VLA for robot manipulation