---
library_name: lerobot
tags:
- act
- robotics
- lerobot

---

# act_block_stack_20k

LeRobot ACT policy for `block_stack` on SO101 data.

| | |
|---|---|
| **Policy** | ACT (`policy.type=act`) |
| **Dataset** | [villekuosmanen/armnetbench_block_stack](https://huggingface.co/datasets/villekuosmanen/armnetbench_block_stack) |
| **Task** | `block_stack` |
| **Action dim** | 6 (single-arm) |
| **Cameras** | `top`, `wrist`, `front` |
| **Training** | 20k total: 10k scratch + 10k warm-start @ bs16 on GCloud A100 (phase-1: pravsels/act_block_stack) |
| **Phase-1 checkpoint** | [pravsels/act_block_stack](https://huggingface.co/pravsels/act_block_stack) |
| **W&B project** | [act_block_stack](https://wandb.ai/pravsels/act_block_stack) |
| **W&B run** | [uk5gu3wt](https://wandb.ai/pravsels/act_block_stack/runs/uk5gu3wt) |

## Checkpoints

The final checkpoint (local step `010000`, 20k total steps) lives at the **repository root** for direct loading.

## Usage

```python
from lerobot.policies.act.modeling_act import ACTPolicy
policy = ACTPolicy.from_pretrained("pravsels/act_block_stack_20k")
```