ACT Policy for PickPlace Task
Trained with LeRobot on TrebleY/pickplace_50_20260902_213020 dataset.
Model Details
- Policy Type: ACT (Action Chunking with Transformers)
- Vision Backbone: ResNet18
- Chunk Size: 100
- Action Steps: 100
- Model Dimension: 512
- Total Params: 51.6M
Training Details
- Total Steps: 10000
- Batch Size: 32
- Optimizer: AdamW
- Learning Rate: 1e-5 (effective)
- Final Loss: 0.174
- L1 Loss: 0.158
- KLD Loss: 0.002
- Training Duration: ~2.6 hours
Dataset
- Dataset:
TrebleY/pickplace_50_20260902_213020 - Episodes: 50
- Total Frames: 20981
- Image Transforms: Enabled
Usage
python
from lerobot.common.policies.act.modeling_act import ACTPolicy
policy = ACTPolicy.from_pretrained("TrebleY/act_pickplace_50")
- Downloads last month
- 6