---
license: mit
tags:
  - robotics
  - isaac-lab
  - xarm
  - teleoperation
---

# Residual Copilot — Data

Teleoperation demonstration data for [Residual Shared Autonomy](https://github.com/shuosha/Residual_Shared_Autonomy). Used by the kNN pilot for nearest-neighbor action retrieval and for sampling episode initial states during environment resets.

## Contents

```
teleop/
├── gearmesh_train_data.npy               # GearMesh teleop trajectories
├── peginsert_train_data.npy              # PegInsert teleop trajectories
├── nutthread_train_data.npy              # NutThread teleop trajectories
├── *_train_data_aug.npy                  # Augmented versions (with noise/perturbations)
└── *_train_data_aug_meta.json            # Augmentation metadata
```

Each `.npy` file is a dictionary keyed by episode ID. Each episode contains `obs.fingertip_pos`, `obs.fingertip_quat`, `obs.fingertip_pos_rel_fixed`, `obs.fingertip_pos_rel_held`, and action trajectories.

For DiffusionPolicy BC training, use the separate LeRobot-format datasets instead (e.g., `shashuo0104/0126_gearmesh_expert_2000`).

## Usage

Data is auto-downloaded at runtime via `resolve_hf()`. See [the main repo](https://github.com/shuosha/Residual_Shared_Autonomy) for setup instructions.

## Related

| Repo | Type | Contents |
|------|------|----------|
| [residual_copilot_assets](https://huggingface.co/datasets/shashuo0104/residual_copilot_assets) | dataset | Robot URDF/USD and object meshes |
| [residual_copilot_models](https://huggingface.co/shashuo0104/residual_copilot_models) | model | BC pilot + RL residual copilot checkpoints |