| library_name: lerobot | |
| tags: | |
| - lerobot | |
| - robotics | |
| - imitation-learning | |
| - isaac-sim | |
| - leisaac | |
| - sim2sim | |
| datasets: | |
| - Aikwed/pick_orange_joint_limited | |
| # GR00T Pick Orange (LeIsaac Sim2Sim) | |
| This repository contains the complete public LeRobot inference checkpoint for a GR00T N1.7 3B policy fine-tuned on [`Aikwed/pick_orange_joint_limited`](https://huggingface.co/datasets/Aikwed/pick_orange_joint_limited). | |
| The policy predicts six SO-101 joint-position actions from the robot state, a wrist RGB camera, a front RGB camera, and the task instruction. It was trained and evaluated for the Pick Orange task in LeIsaac using a simulation-to-simulation (sim2sim) workflow. | |
| ## Scope and limitations | |
| - Robot: SO-101 Follower | |
| - Simulator: LeIsaac / Isaac Sim | |
| - Task: Pick Orange | |
| - Evaluation setting: sim2sim | |
| - Real-robot deployment has not been validated. | |
| The public `model.safetensors` file contains the model weights. The processor JSON and safetensors files are required for inference with the matching LeRobot version. | |