rskill-act-libero

OpenRAL rSkill — Action Chunking Transformer (Zhao et al., 2023) fine-tuned on HuggingFaceVLA/libero, packaged for the LIBERO Franka-Panda embodiment.

Upstream model

Field	Value
Weights	hf://Deepkar/libero-test-act
Architecture	ResNet-18 backbone · 4+1 encoder/decoder · latent VAE · chunk_size=100
Action	7-D delta-EEF + gripper
Dataset	HuggingFaceVLA/libero (Apache-2.0)
License	Apache-2.0
Paper	arxiv:2304.13705 — Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware

Supported robots

Robot	Embodiment tag	Status	Notes
Franka Panda (LIBERO sim)	franka_panda (LIBERO embodiment tag)	✓ sim	Native training target; closed-loop rollout reaches is_success=True on libero_spatial/2 in ~91 steps on a single seed.

Sensors required

Key	Modality	Resolution	Format
observation.images.image	RGB	256 × 256	float32
observation.images.image2	RGB	256 × 256	float32
observation.state	proprioception	(8,)	float32 (LIBERO Franka layout: pos3 + axisangle3 + grip2)

Manifest summary

Field	Value
name	OpenRAL/rskill-act-libero
version	0.1.0
license	apache-2.0
role	s1
embodiment_tags	franka_panda
runtime / quantization.dtype	pytorch / fp32
weights_uri	hf://Deepkar/libero-test-act
chunk_size	100
commercial_use_allowed	true

Full schema: openral_core.schemas.RSkillManifest.

Run

CC=/usr/bin/gcc uv sync --group libero      # first time only
openral sim run --config scenes/benchmark/libero_spatial.yaml --rskill rskills/act-libero \
           --rskill rskills/act-libero

The shipped sim YAML pins libero_spatial/0 for a 200-step single-episode rollout. Sweep tasks with --task libero_spatial/<n>. A spot-check on libero_spatial/2 reaches is_success=True in ~91 steps (reward 1.0) on a single seed.

Camera & state contract

LIBERO emits images={"camera1": agentview, "camera2": eye_in_hand} while this checkpoint's input features are observation.images.image / observation.images.image2. The manifest's image_preprocessing block rewrites the batch keys at step time:

image_preprocessing:
  flip_180: true            # HuggingFaceVLA/libero is captured rotated 180°
  aliases:
    camera1: image
    camera2: image2

The state_contract.dim: 8 declaration confirms the proprio width. Because the upstream training set is HuggingFaceVLA/libero — the same dataset the smolvla / pi05 / xvla LIBERO checkpoints in this repo were finetuned on — the state semantics (pos3 + axisangle3 + grip2) line up with OpenRAL's LIBERO backend end-to-end, with no quat-vs-axisangle mismatch.

Benchmarks

None measured yet. Populate eval/ with openral benchmark run JSON fixtures before publishing a headline number.

License

Apache-2.0 — both the wrapping rSkill package (rskill.yaml, README.md) and the wrapped upstream weights at hf://Deepkar/libero-test-act. Commercial use is allowed (commercial_use_allowed: true).