lerobot/being_h05_robocasa

Being-H0.5 RoboCasa — LeRobot conversion

This is a deterministic LeRobot-format conversion of BeingBeyond/Being-H05-2B_robocasa. It contains the policy configuration, serialized preprocessor and postprocessor pipelines, tokenizer, conversion report, and safetensors weights.

Revisions

• Author source: ab1204954cb5e68f0bc52a376982e962c763bbbe
• Source checkpoint: 75989b121eab7d0a8e421d846acf268c2938883d
• LeRobot integration: 27152836dab56f5e6b1cecd3fa62f45bbd4e274d

Validation

• 990 source tensors mapped to 1,086 native-layout tensors
• Native LeRobot policy using Transformers Qwen3 primitives and InternVLVisionModel
• 0 missing, unexpected, duplicate, or shape-mismatched tensors
• Strict offline save/reload passed
• Native Qwen3-MoT, flow training, four-step sampling, and active MPG refinement match the author implementation bitwise in CUDA BF16
• SHA-256: bbe5c8d264e705296851b846e625e936fb1bf05f92b39d311777cb59d0e57aa1
• Author-vs-LeRobot model forward pass at batch sizes 1 and 2: maximum absolute error 0

The checkpoint preserves Being-H0.5's fixed semantic 200D state/action slots, validity masks, three-camera processing, raw task string, and four-step flow-matching sampler. It consumes the named state and action fields from the released model contract; an environment-specific vector adapter is not included.

Usage

Install LeRobot with the Being-H0.5 extra:

pip install -e ".[being_h05]"

No author-source checkout or PYTHONPATH change is required. Load the checkpoint through the standard LeRobot policy interface after providing the required named state fields:

--policy.path=lerobot/being_h05_robocasa