3d6b73753a
- Add pick_test_tube task: USDC asset repackaging, grasp generation, task config - Add tools: usdc_to_obj.py, repackage_test_tube.py, fix_test_tube_materials.py - Add custom_task_guide.md: full Chinese documentation for creating custom tasks - Add crawled InternDataEngine online docs (23 pages) - Add grasp generation script (gen_tube_grasp.py) and pipeline config
83 lines
3.5 KiB
Markdown
83 lines
3.5 KiB
Markdown
# Source: https://internrobotics.github.io/InternDataEngine-Docs/policy/training.html
|
||
|
||
# Training [](#training)
|
||
|
||
This guide covers data format conversion and policy training for validating generated simulation data.
|
||
|
||
## Part 1: LMDB to LeRobot Data Conversion [](#part-1-lmdb-to-lerobot-data-conversion)
|
||
|
||
The simulation data generated by InternDataEngine is stored in LMDB format. To use this data for policy training, you need to convert it to LeRobot format.
|
||
|
||
### Step 1: Install LeRobot v2.1 [](#step-1-install-lerobot-v2-1)
|
||
|
||
We use LeRobot v2.1 format for data storage. Install the LeRobot 2.1 repo.
|
||
|
||
### Step 2: Convert LMDB to LeRobot v2.1 [](#step-2-convert-lmdb-to-lerobot-v2-1)
|
||
|
||
Use the conversion scripts in ``[policy/lmdb2lerobotv21](https://github.com/InternRobotics/InternDataEngine/tree/master/policy/lmdb2lerobotv21)directory.
|
||
|
||
We provide conversion scripts for different robot platforms:
|
||
|
||
- **lmdb2lerobot_lift2_a1.py **( script ): Lift2 (ARX).
|
||
- **lmdb2lerobot_split_aloha_a1.py **( script ): Split Aloha.
|
||
- **lmdb2lerobot_genie1_a1.py **( script ): Genie1.
|
||
- **lmdb2lerobot_franka_a1.py **( script ): Franka FR3.
|
||
- **lmdb2lerobot_frankarobotiq_a1.py **( script ): Franka with Robotiq gripper.
|
||
|
||
Example usage:
|
||
bash
|
||
```
|
||
python lmdb2lerobot_lift2_a1.py \
|
||
--src_path ${src_path} \
|
||
--save_path ${save_path} \
|
||
--repo_id ${repo_id} \
|
||
--num-threads ${num_threads} \
|
||
--num_demos ${num_demos}
|
||
```
|
||
1
|
||
2
|
||
3
|
||
4
|
||
5
|
||
6
|
||
|
||
**Parameters: **
|
||
|
||
- **--src_path **( str ): Path to the source LMDB data directory.
|
||
- **--save_path **( str ): Path to save the converted LeRobot dataset.
|
||
- **--repo_id **( str ): Dataset repository identifier.
|
||
- **--num-threads **( int ): Number of threads for parallel processing.
|
||
- **--num_demos **( int ): Number of demonstrations to convert (optional).
|
||
|
||
### Step 3: Convert to LeRobot v3.0 (Optional) [](#step-3-convert-to-lerobot-v3-0-optional)
|
||
|
||
If you need LeRobot v3.0 format for training, please install LeRobot 3.0. Then use the conversion script:
|
||
bash
|
||
```
|
||
python convertv21_to_v30.py --input_path ${v21_path} --output_path ${v30_path}
|
||
```
|
||
1
|
||
|
||
The conversion code is available at ``[policy/lmdb2lerobotv21/convertv21_to_v30.py](https://github.com/InternRobotics/InternDataEngine/tree/master/policy/lmdb2lerobotv21/convertv21_to_v30.py).
|
||
|
||
## Part 2: Policy Training with π 0 [](#part-2-policy-training-with-π0)
|
||
|
||
As described in the [InternData-A1 paper](https://arxiv.org/pdf/2511.16651), we used multi-machine, multi-GPU JAX-based π 0 for data validation.
|
||
|
||
We have implemented a JAX-based, multi-nodes, multi-GPU training pipeline that supports multi-dataset mixed training for π 0 .
|
||
|
||
### Features [](#features)
|
||
|
||
- **Multi-machine, multi-GPU training **: Scale training across multiple nodes
|
||
- **Multi-dataset mixed training **: Train on multiple datasets simultaneously
|
||
- **JAX-based implementation **: High-performance training with JAX/Flax
|
||
|
||
### Installation, Training, and Deployment [](#installation-training-and-deployment)
|
||
|
||
For detailed instructions on installation, training, and deployment, please refer to the [openpi-InternData-A1 README](https://github.com/InternRobotics/InternDataEngine/blob/master/policy/openpi-InternData-A1/README.md).
|
||
|
||
## References [](#references)
|
||
|
||
- [LeRobot](https://github.com/huggingface/lerobot)- HuggingFace LeRobot
|
||
- [InternData-A1 Paper](https://arxiv.org/pdf/2511.16651)- InternData-A1: A High-Fidelity Synthetic Data Generator for Robotic Manipulation
|
||
- [openpi-InternData-A1](https://github.com/InternRobotics/InternDataEngine/blob/master/policy/openpi-InternData-A1/)- JAX-based π 0 training code |