maple.backend.policy.gr00tn15.GR00TN15Policy
- class maple.backend.policy.gr00tn15.GR00TN15Policy
Backend for NVIDIA Isaac GR00T N1.5/N1.6 vision-language-action models.
GR00T is an open foundation model for generalized humanoid robot reasoning and skills. It is a cross-embodiment model that takes multimodal input including language, images, and proprioception to perform manipulation tasks in diverse environments.
Key features: - Flow matching transformer for action prediction - Cross-embodiment support via EmbodimentTag system - Multi-camera view support - Proprioceptive state conditioning - Action chunk prediction (default horizon: 16)
The backend manages Docker containers running the GR00T inference server, which loads the model from HuggingFace and serves predictions via HTTP API.
Model Load Kwargs: - embodiment_tag: Robot embodiment identifier (e.g., ‘GR1’, ‘NEW_EMBODIMENT’) - data_config: Data configuration name (e.g., ‘fourier_gr1_arms_only’, ‘libero’) - denoising_steps: Number of flow matching denoising steps (default: 4)
Methods
act(handle, payload, instruction[, model_kwargs])Get action prediction for a single observation.
info()Get policy backend information and capabilities.
Attributes