admin/lerobot
1
0
Fork 0
mirror of https://github.com/huggingface/lerobot.git synced 2026-05-21 19:49:49 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix styling

Browse Source
This commit is contained in:
Jade Choghari
2025-11-26 15:34:45 +01:00
parent fbcf118dcb
commit 863ae89ff2
1 changed files with 29 additions and 22 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docs/source/xvla.mdx
+20 -13
View File
@@ -130,12 +130,12 @@ lerobot-train \
### Training Parameters Explained ### Training Parameters Explained
| Parameter | Default | Description | | Parameter | Default | Description |
|-----------|---------|-------------| | -------------------------- | ------- | ---------------------------------------- |
| `freeze_vision_encoder` | `True` | Freeze the VLM vision encoder weights | | `freeze_vision_encoder` | `True` | Freeze the VLM vision encoder weights |
| `freeze_language_encoder` | `True` | Freeze the VLM language encoder weights | | `freeze_language_encoder` | `True` | Freeze the VLM language encoder weights |
| `train_policy_transformer` | `True` | Allow policy transformer layers to train | | `train_policy_transformer` | `True` | Allow policy transformer layers to train |
| `train_soft_prompts` | `True` | Allow soft prompts to train | | `train_soft_prompts` | `True` | Allow soft prompts to train |
**💡 Best Practice**: For Phase II adaptation to new embodiments, freeze the VLM encoders and only train the policy transformer and soft prompts. This provides excellent sample efficiency with minimal compute. **💡 Best Practice**: For Phase II adaptation to new embodiments, freeze the VLM encoders and only train the policy transformer and soft prompts. This provides excellent sample efficiency with minimal compute.
@@ -167,13 +167,13 @@ X-VLA uses an **Action Registry** system to handle different action spaces and e
#### Available Action Modes #### Available Action Modes
| Action Mode | Action Dim | Description | Use Case | | Action Mode | Action Dim | Description | Use Case |
|-------------|------------|-------------|----------| | ---------------- | --------------------- | ------------------------------------------- | ------------------------------------ |
| `ee6d` | 20 | End-effector with xyz, 6D rotation, gripper | Dual-arm setups with spatial control | | `ee6d` | 20 | End-effector with xyz, 6D rotation, gripper | Dual-arm setups with spatial control |
| `joint` | 14 | Joint-space with gripper | Direct joint control robots | | `joint` | 14 | Joint-space with gripper | Direct joint control robots |
| `agibot_ee6d` | 20 | AGI-bot variant with MSE loss | AGI-bot platforms | | `agibot_ee6d` | 20 | AGI-bot variant with MSE loss | AGI-bot platforms |
| `franka_joint7` | 7 | Franka Panda 7-joint control | Franka robots without gripper | | `franka_joint7` | 7 | Franka Panda 7-joint control | Franka robots without gripper |
| `so101_bimanual` | 20 (model), 12 (real) | SO101 bimanual robot | Bimanual manipulation tasks | | `so101_bimanual` | 20 (model), 12 (real) | SO101 bimanual robot | Bimanual manipulation tasks |
#### Why Action Modes Matter #### Why Action Modes Matter
@@ -381,6 +381,7 @@ preprocessor = PolicyProcessorPipeline(
When your dataset has fewer action dimensions than the pretrained model: When your dataset has fewer action dimensions than the pretrained model:
**Option 1**: Use padding (automatic in most action modes) **Option 1**: Use padding (automatic in most action modes)
```python ```python
# Model expects 20D, dataset has 12D # Model expects 20D, dataset has 12D
# Action mode handles padding internally # Action mode handles padding internally
@@ -388,6 +389,7 @@ action_mode = "so101_bimanual" # Pads 12 → 20
``` ```
**Option 2**: Create a custom action mode that maps dimensions explicitly **Option 2**: Create a custom action mode that maps dimensions explicitly
```python ```python
@register_action("my_mapped_action") @register_action("my_mapped_action")
class MappedActionSpace(BaseActionSpace): class MappedActionSpace(BaseActionSpace):
@@ -406,15 +408,19 @@ class MappedActionSpace(BaseActionSpace):
### Common Issues ### Common Issues
**Issue**: "Action dimension mismatch" **Issue**: "Action dimension mismatch"
- **Solution**: Check that your `action_mode` matches your robot's action space. Create a custom action mode if needed. - **Solution**: Check that your `action_mode` matches your robot's action space. Create a custom action mode if needed.
**Issue**: "Image values outside [0, 1] range" **Issue**: "Image values outside [0, 1] range"
- **Solution**: Ensure images are preprocessed with `XVLAImageToFloatProcessorStep` before normalization. - **Solution**: Ensure images are preprocessed with `XVLAImageToFloatProcessorStep` before normalization.
**Issue**: "Domain ID not found" **Issue**: "Domain ID not found"
- **Solution**: Make sure `XVLAAddDomainIdProcessorStep` is in your preprocessing pipeline with the correct domain_id. - **Solution**: Make sure `XVLAAddDomainIdProcessorStep` is in your preprocessing pipeline with the correct domain_id.
**Issue**: "Low success rate on new embodiment" **Issue**: "Low success rate on new embodiment"
- **Solution**: - **Solution**:
1. Verify your action_mode is correct 1. Verify your action_mode is correct
2. Check that soft prompts are being trained (`train_soft_prompts=True`) 2. Check that soft prompts are being trained (`train_soft_prompts=True`)
@@ -422,6 +428,7 @@ class MappedActionSpace(BaseActionSpace):
4. Consider increasing training steps 4. Consider increasing training steps
**Issue**: "Out of memory during training" **Issue**: "Out of memory during training"
- **Solution**: - **Solution**:
1. Reduce `chunk_size` (e.g., from 32 to 16) 1. Reduce `chunk_size` (e.g., from 32 to 16)
2. Enable gradient checkpointing 2. Enable gradient checkpointing