---
license: apache-2.0
library_name: lerobot
pipeline_tag: robotics
tags:
  - flow-matching
  - diffusion-policy
  - rectified-flow
  - lerobot
  - so101
  - leisaac
  - pick-orange
  - isaac-sim
base_model: []
datasets:
  - LightwheelAI/leisaac-pick-orange
language:
  - en
---

# FlowHeads-DiffusionPolicy-PickOrange (DP-FlowHead)

**DP-FlowHead** — a [LeRobot](https://github.com/huggingface/lerobot) Diffusion-Policy
conv-UNet (ResNet-18 + SpatialSoftmax + FiLM 1D-conv UNet, ~267M) whose **DDPM
denoising head is replaced by a rectified-flow head**: predict the straight-line
velocity `v = x1 − x0` and sample by **Euler ODE** integration (NFE=10). Trained
**from scratch** on [LeIsaac SO-101 PickOrange](https://github.com/LightwheelAI/leisaac)
(single arm, 2 RGB cams, 60 demos). This ckpt = **step-9800 ≈ 4.3 epoch (best)**.

针对 [LeIsaac SO-101 PickOrange](https://github.com/LightwheelAI/leisaac) **从头训练**的
LeRobot 策略:在 Diffusion-Policy 的 conv-UNet 骨干上,把 **DDPM 去噪头换成 rectified-flow 头**
(速度场 `v=x1−x0` + Euler 10 步积分)。本 ckpt = **step-9800 ≈ 4.3 epoch(best)**。

<video controls width="640" src="https://huggingface.co/wsagi/FlowHeads-DiffusionPolicy-PickOrange/resolve/main/demo.mp4"></video>

> _Closed-loop demo in Isaac Sim — SO-101 picking oranges into the plate._

## 🔗 Project repos / 项目仓库

- [**vitorcen/FlowHeads**](https://github.com/vitorcen/FlowHeads) — the flow-matching action-head umbrella (this model's code: `flowdp/`)
- [**vitorcen/LeIsaac-Training**](https://github.com/vitorcen/LeIsaac-Training) — the LeIsaac PickOrange benchmark + eval harness
- [**vitorcen/isaaclab-experience**](https://github.com/vitorcen/isaaclab-experience) — Isaac Lab multi-policy umbrella (parent project)

## Results — strict 20-round (PickOrange, closed-loop Isaac)

**Headed, 20 rounds (60 episodes), `EPISODE_LENGTH_S=120`, `MAX_ROUND_WALL_S=180`, h=8.**

| metric | value |
|---|---|
| E(🍊)/ep | **1.35 / 3 = 45.0 %** (27/60) |
| P(3) — full round (all 3) | 20 % (4/20) |
| P(≥2) | 40 % |
| avg episode | 171 s |
| 20-ep raw | `[3,1,2,3,1,2,1,0,3,1,0,2,2,1,1,1,3,0,0,0]` |

**45.0 % ties the strongest baseline ACT (43.3 %)** on this 60-demo task (a single
20-round run carries ±~9 %, so this is a tie, not a win). Notably, an earlier
5-round *quick* eval with a tight 90 s wall-cap had **mis-scored this policy near zero**:
DP-FlowHead is slow (successful episodes take ~136 s) and the 90 s cap truncated its
completions. The strict 180 s eval is what reveals the real 45 %. See the
[FlowHeads architecture × objective study](https://github.com/vitorcen/FlowHeads) for the
full {conv-UNet, DiT} × {DDPM, flow} matrix.

## Usage

```python
from lerobot.policies.factory import get_policy_class
# requires the FlowHeads package on PYTHONPATH so "flowdp" is registered:
#   pip install -e .  # from https://github.com/vitorcen/FlowHeads
policy = get_policy_class("flowdp").from_pretrained("wsagi/FlowHeads-DiffusionPolicy-PickOrange")
```

Config: `type=flowdp`, `n_action_steps=8`, `num_inference_steps=10` (Euler NFE).
Closed-loop scoring uses the LeIsaac benchmark harness
(`scripts/benchmark/run_one.sh`, eval-side lerobot 0.4.x).