lerobot/docs/source/hil_collection.mdx

# Human In the Loop: Recovery and Correction Data Collection

RaC (Recovery and Correction) is a human-in-the-loop data collection and training paradigm that improves robot policy performance on long-horizon tasks by explicitly teaching recovery and correction behaviors.

---

## Why RaC? The Problem with Standard Data Collection

### Standard Behavioral Cloning Data Collection Limitations

Standard behavior cloning trains policies on successful demonstrations. This approach can be sensitive to distribution shift and compounding errors. Because during deployment small errors can cascade and push the robot into states never seen during training.
This is where RaC which builds on work like Dagger and HG-DAgger comes in.

### Prior Human-in-the-Loop Methods

**DAgger** (Dataset Aggregation) addresses distribution shift by:

- Running the novice policy to collect states
- Querying expert for correct actions at those states
- Aggregating new labels into training set

**HG-DAgger** (Human-Gated DAgger) improves on DAgger by:

- Giving human full control authority during interventions
- Human takes over when unsafe, provides correction, returns control
- Better action labels because human has uninterrupted control

### RaC

RaC explicitly collects **recovery + correction** data:

```
BC/DAgger:   policy → mistake → human corrects → continue
RaC:         policy → mistake → human RECOVERS (teleop back) → CORRECTS → END
```

This Human in the loop approach follows two rules:

**Rule 1 (Recover then Correct)**:

- Every intervention starts with human teleoperating back to an in-distribution state
- Then human provides correction to complete the current subtask
- Both segments are recorded as training data
- This teaches the policy: "when things go wrong, go back and retry"

**Rule 2 (Terminate after Intervention)**:

- Episode ends after correction completes
- Avoids mixed policy/human data on later subtasks

---

## Comparison Table

| Method    | Data Type                       | Recovery Behavior | Correction Behavior |
| --------- | ------------------------------- | ----------------- | ------------------- |
| BC        | Success only                    | ✗                 | ✗                   |
| DAgger    | Success + corrections           | ✗                 | ✓                   |
| HG-DAgger | Success + corrections           | Sometimes         | ✓                   |
| RaC       | Success + recovery + correction | ✓ Explicit        | ✓                   |

---

## Hardware Requirements

### Teleoperator Requirements

The HIL data collection script requires **teleoperators with active motors** that can:

- Enable/disable torque programmatically
- Move to target positions (to mirror robot state when pausing)

**Compatible teleoperators:**

- `so101_leader` - SO-101 Leader Arm
- `openarms_mini` - OpenArms Mini (via third-party plugin)

---

## Scripts

Two scripts are provided depending on your policy's inference speed:

| Script                       | Use Case                                   | Models                |
| ---------------------------- | ------------------------------------------ | --------------------- |
| `hil_data_collection.py`     | Standard synchronous inference             | ACT, Diffusion Policy |
| `hil_data_collection_rtc.py` | Real-Time Chunking for high-latency models | Pi0, Pi0.5, SmolVLA   |

---

## The Pipeline

```
┌─────────────────────────────────────────────────────────────────────────┐
│                         RaC Training Pipeline                           │
├─────────────────────────────────────────────────────────────────────────┤
│                                                                         │
│  1. PRE-TRAINING (Standard BC)                                          │
│     └─> Train initial policy on clean demonstrations                    │
│                                                                         │
│  2. HIL DATA COLLECTION (Human-in-the-loop)                             │
│     ├─> Policy runs autonomously                                        │
│     ├─> Human monitors and intervenes when failure imminent             │
│     │   ├─> RECOVERY: Human teleoperates robot back to good state       │
│     │   └─> CORRECTION: Human completes the current subtask             │
│     └─> Episode terminates after correction (Rule 2)                    │
│                                                                         │
│  3. REWARD LABELING (Optional: SARM)                                    │
│     └─> Compute progress rewards for advantage-weighted training        │
│                                                                         │
│  4. FINE-TUNING                                                         │
│     └─> Train on combined demos + HIL data (optionally with RA-BC)      │
│                                                                         │
└─────────────────────────────────────────────────────────────────────────┘
```

---

## Step-by-Step Guide

### Step 1: Pre-train a Base Policy

First, train a policy on your demonstration dataset:

```bash
python src/lerobot/scripts/lerobot_train.py \
    --dataset.repo_id=your-username/demo-dataset \
    --policy.type=pi0 \
    --output_dir=outputs/pretrain \
    --batch_size=32 \
    --steps=50000
```

### Step 2: Collect HIL Data

**Standard inference (ACT, Diffusion Policy):**

```bash
python examples/rac/hil_data_collection.py \
    --robot.type=so100_follower \
    --robot.port=/dev/tty.usbmodem58760431541 \
    --robot.cameras="{ front: {type: opencv, index_or_path: 0, width: 640, height: 480, fps: 30}}" \
    --teleop.type=so100_leader \
    --teleop.port=/dev/tty.usbmodem58760431551 \
    --policy.path=outputs/pretrain/checkpoints/last/pretrained_model \
    --dataset.repo_id=your-username/hil-dataset \
    --dataset.single_task="Pick up the cube and place it in the bowl" \
    --dataset.num_episodes=50
```

**With RTC for large models (Pi0, Pi0.5, SmolVLA):**

For models with high inference latency, use the RTC script for smooth execution:

```bash
python examples/rac/hil_data_collection_rtc.py \
    --robot.type=so100_follower \
    --teleop.type=so100_leader \
    --policy.path=outputs/pretrain/checkpoints/last/pretrained_model \
    --dataset.repo_id=your-username/hil-rtc-dataset \
    --dataset.single_task="Pick up the cube" \
    --rtc.execution_horizon=20 \
    --interpolation=true
```

**Controls (Keyboard + Foot Pedal):**

| Key / Pedal                | Action                                             |
| -------------------------- | -------------------------------------------------- |
| **SPACE** / Right pedal    | Pause policy (teleop mirrors robot, no recording)  |
| **c** / Left pedal         | Take control (start correction, recording resumes) |
| **→** / Right pedal        | End episode (save) - when in correction mode       |
| **←**                      | Re-record episode                                  |
| **ESC**                    | Stop session and push to hub                       |
| Any key/pedal during reset | Start next episode                                 |

**The HIL Protocol:**

1. Watch the policy run autonomously (teleop is idle/free)
2. When you see imminent failure, press **SPACE** or **right pedal** to pause
   - Policy stops
   - Teleoperator moves to match robot position (torque enabled)
   - No frames recorded during pause
3. Press **c** or **left pedal** to take control
   - Teleoperator torque disabled, free to move
   - **RECOVERY**: Teleoperate back to a good state
   - **CORRECTION**: Complete the subtask
   - All movements are recorded
4. Press **→** or **right pedal** to save and end episode
5. **RESET**: Teleop moves to robot position, you can move robot to starting position
6. Press any key/pedal to start next episode

The recovery and correction segments teach the policy how to recover from errors.

**Foot Pedal Setup (Linux):**

If using a USB foot pedal (PCsensor FootSwitch), ensure access:

```bash
sudo setfacl -m u:$USER:rw /dev/input/by-id/usb-PCsensor_FootSwitch-event-kbd
```

### Step 3: (Optional) Compute SARM Rewards

For advantage-weighted training (RA-BC / Pi0.6-style), compute SARM progress values:

```bash
python src/lerobot/policies/sarm/compute_rabc_weights.py \
    --dataset-repo-id your-username/hil-dataset \
    --reward-model-path your-username/sarm-model \
    --head-mode sparse \
    --push-to-hub
```

### Step 4: Fine-tune Policy

Fine-tune on the HIL data:

```bash
# Without RA-BC (standard fine-tuning)
python src/lerobot/scripts/lerobot_train.py \
    --dataset.repo_id=your-username/hil-dataset \
    --policy.type=pi0 \
    --policy.pretrained_path=outputs/pretrain/checkpoints/last/pretrained_model \
    --output_dir=outputs/hil_finetune \
    --steps=20000

# With RA-BC (advantage-weighted, Pi0.6-style)
python src/lerobot/scripts/lerobot_train.py \
    --dataset.repo_id=your-username/hil-dataset \
    --policy.type=pi0 \
    --policy.pretrained_path=outputs/pretrain/checkpoints/last/pretrained_model \
    --output_dir=outputs/hil_finetune_rabc \
    --use_rabc=true \
    --rabc_kappa=0.01 \
    --steps=20000
```

---

## Connection to Pi0.6 / RECAP

Pi0.6's RECAP method shares similar principles:

- Collect autonomous rollouts + expert interventions
- Use value function to compute **advantages**: A(s,a) = V(s') - V(s)
- **Advantage conditioning**: Weight training based on expected improvement

In LeRobot, we can use **SARM** as the value function:

- SARM progress φ(s) ∈ [0,1] measures task completion
- Progress delta = φ(s') - φ(s) approximates advantage
- RA-BC uses these to weight training samples (higher weight for good corrections)

---

## Tips for Effective HIL Collection

### When to Intervene

Intervene when you see:

- Robot about to make an irreversible mistake
- Robot hesitating or showing uncertain behavior
- Robot deviating from expected trajectory

### Recovery: Teleoperating Back to Good State

During recovery, teleoperate the robot back to a state where:

- The robot is in a familiar, in-distribution configuration
- The current subtask can still be completed
- The recovery trajectory itself is informative training data

### Quality of Corrections

During correction:

- Provide **confident, clean** trajectories
- Complete the current subtask fully
- Don't overcorrect or add unnecessary movements

---

## Iterative Improvement

HIL data collection can be applied iteratively:

```
┌─────────────────────────────────────────────────────────────────────────┐
│  Policy v0 (demos)                                                      │
│       ↓                                                                 │
│  HIL Collection (target current failure modes) → Policy v1              │
│       ↓                                                                 │
│  HIL Collection (target new failure modes) → Policy v2                  │
│       ↓                                                                 │
│  ... (repeat until satisfactory performance)                            │
└─────────────────────────────────────────────────────────────────────────┘
```

---

## References

```bibtex
@article{hu2025rac,
  title={RaC: Robot Learning for Long-Horizon Tasks by Scaling Recovery and Correction},
  author={Hu, Zheyuan and Wu, Robyn and Enock, Naveen and Li, Jasmine and Kadakia, Riya and Erickson, Zackory and Kumar, Aviral},
  journal={arXiv preprint arXiv:2509.07953},
  year={2025}
}

@article{kelly2019hgdagger,
  title={HG-DAgger: Interactive Imitation Learning with Human Experts},
  author={Kelly, Michael and Sidrane, Chelsea and Driggs-Campbell, Katherine and Kochenderfer, Mykel J},
  journal={arXiv preprint arXiv:1810.02890},
  year={2019}
}

@article{pi2025recap,
  title={π∗0.6: a VLA That Learns From Experience},
  author={Physical Intelligence},
  year={2025}
}

@article{chen2025sarm,
  title={SARM: Stage-Aware Reward Modeling for Long Horizon Robot Manipulation},
  author={Chen, Qianzhong and Yu, Justin and Schwager, Mac and Abbeel, Pieter and Shentu, Yide and Wu, Philipp},
  journal={arXiv preprint arXiv:2509.25358},
  year={2025}
}
```