SmolVLA orange-picking policy for LeIsaac

edge-inference publishes the checkpoint with task-specific training and evaluation instructions. The card names LeIsaac-SO101-PickOrange-v0 and distinguishes main and single-rank training branches. Its generic Hub usage snippet differs from the task-specific instructions; start with the latter.

The task-specific card reports LeRobot v0.4.5 and requires LeIsaac plus Isaac Sim. Match the checkpoint branch, observation keys and policy processors to the task; physical SO101 transfer is unverified.

Sources checked 2026-09-21. Shared by a community member; original work by edge-inference.

• Source: https://huggingface.co/edge-inference/smolvla-so101-pick-orange
• Source: https://github.com/LightwheelAI/leisaac Known limits:
• Immutable resource revisions have not been captured; linked default branches/cards can change.
• No independent installation, policy evaluation or hardware reproduction was performed.
• Checkpoint commit, exact simulator asset versions and independent evaluation results are unknown.
• Compatibility with physical hardware has not been established.

Restored from the original local preview catalog; original creator attribution and reviewed source notes are preserved.