runwayml/praxis-1-droid

Praxis-1 DROID

Praxis-1 policy fine-tuned for the single-arm DROID rig (Franka + Robotiq gripper).

Model	27.2B-parameter MoE video-diffusion policy: chunk-causal trunk + 0.3B action expert
Frozen dependencies (not in this repo)	Wan2.2 VAE, Qwen/Qwen3.5-2B text encoder
Embodiment	single arm, cameras exterior + wrist
Control	15 Hz, 24-step action chunks
Action space	10-D [xyz(3) rot6d(6) grip(1)], relative to the chunk-start pose
Weights	bf16, 54.4 GB, 17 safetensors shards (model.safetensors.index.json)

Normalization statistics and the action codec are in config.json; no dataset statistics are needed at inference time.

Files

• config.json: praxis model config (model_type: praxis, format_version: 2)
• model-000NN-of-00017.safetensors + model.safetensors.index.json: sharded weights
• SHA256SUMS: checksums for every file above

Usage

from huggingface_hub import snapshot_download

path = snapshot_download("runwayml/praxis-1-droid")

Load path with runwayml/praxis-1-oss (serve it with configs/serve_droid.yaml) or with the LeRobot praxis policy via runwayml/praxis-1-droid-lerobot. Inference needs one 80 GB GPU. To adapt it to your own DROID-style data on one 80 GB GPU, see "LoRA on praxis-1-droid" in praxis/train/README.md.

On the DROID autoregressive prompt-adherence set of runwayml/praxis-1-eval, this checkpoint reaches 52.5 % full adherence (95 % CI [35.8, 68.3]).

Related

• runwayml/praxis-1-base: base model, no action heads; the start of a fine-tune for a new rig
• runwayml/praxis-1-droid: single-arm DROID policy
• runwayml/praxis-1-droid-lerobot: praxis-1-droid in the LeRobot policy layout
• runwayml/praxis-1-yam: bimanual YAM policy
• runwayml/praxis-1-eval: evaluation sets
• runwayml/praxis-1-oss: reference code (inference, serving, fine-tuning)

License

Apache License 2.0. The frozen dependencies (Wan2.2 VAE, Qwen/Qwen3.5-2B) are distributed separately under their own licenses.