---
license: apache-2.0
library_name: lerobot
tags:
- lerobot
- diffusion-policy
- imitation-learning
- robotics
- ur3e
- bimanual
datasets:
- WetLabRoboData/lerobot-data-needle
pipeline_tag: robotics
base_model: WetLabRoboData/diffusion-multitask_12task_mix-multitask
---

# diffusion-needle-finetuned

LeRobot diffusion policy for the **needle** task.

| Field              | Value |
|--------------------|-------|
| Policy family      | diffusion (LeRobot) |
| Variant            | Finetuned (multitask pretrain, then finetuned on this task's finetune pool) |
| Training dataset   | [`WetLabRoboData/lerobot-data-needle`](https://huggingface.co/datasets/WetLabRoboData/lerobot-data-needle) |
| Base model         | [`WetLabRoboData/diffusion-multitask_12task_mix-multitask`](https://huggingface.co/WetLabRoboData/diffusion-multitask_12task_mix-multitask) |
| Target task        | needle |
| Eval episodes      | 20 |
| Eval successes     | 15 / 20 |
| Robot              | UR3e bimanual (3 cameras) |

## Evaluation

Rollout videos + per-episode outcomes: [`WetLabRoboData/eval-diffusion-needle-finetuned`](https://huggingface.co/datasets/WetLabRoboData/eval-diffusion-needle-finetuned).

## Load and deploy

```python
from lerobot.policies.diffusion.modeling_diffusion import DiffusionPolicy
policy = DiffusionPolicy.from_pretrained("WetLabRoboData/diffusion-needle-finetuned")
```

## Provenance

Reorganized from `WetLabRoboData/lerobot-data-rama-lbm_finetune_needle` on 2026-10-04. The archived
training outputs (checkpoints, train_config.json, wandb/) are preserved in
the `old/` subfolder of that source repo for traceability.