README.md

---
license: cc-by-nc-4.0
pretty_name: HTT Dataset & Model
tags:
- robotics
- tactile-sensing
- multimodal
size_categories:
- 1M<n<10M
---

# HTT Dataset & Model

Paired multi-sensor tactile data (**1.59M frames**) **and the pretrained
model checkpoint** for the Heterogeneous Tactile Transformer (HTT). Four
sensors — two vision-based, two taxel arrays — organized into four task
splits.

| sensor | type | raw format |
|---|---|---|
| `gsmini` | vision (GelSight Mini) | JPEG `[224, 224, 3]` |
| `9dtact` | vision (9DTact) | JPEG `[224, 224, 3]` |
| `xela` | taxel array (Xela uSkin) | float `[T, 72]` |
| `tac02` | taxel array (TAC-02) | float `[T, 66]` |

```
├── pretrain/        1. self-supervised pretraining (paired episodes, fully unlabeled)
├── classification/  2. 20-object classification (paired episodes, labeled)
├── force/           3. 6D force estimation (static probe episodes, aligned F/T)
├── slip/            4. slip-stage detection (sliding episodes, 3-class labels)
├── bg_data/         background/reference frames per sensor
├── htt_4sensors_best.pth  pretrained HTT checkpoint (~69 MB)
├── STATS.json       measured per-split statistics
├── FORMAT.md        detailed per-split format documentation
└── MANIFEST.sha256  checksums for every dataset file
```

## Model checkpoint

This repo is the single home for **dataset & model**. The pretrained HTT
checkpoint (`htt_4sensors_best.pth`, ~69 MB, SHA-256
`024f4c3a067168197d0a6996bbca5c03e744ed5abd1d35a666dbf78e7ac673f0`):

```bash
hf download AllenBi21/HTT-dataset htt_4sensors_best.pth --repo-type dataset --local-dir checkpoints
```

Inference API and training code: [github.com/jxbi1010/HTT](https://github.com/jxbi1010/HTT).

## Statistics

One *paired sample* = 1 optical frame + 1 synchronized chunk of 10 taxel
frames; npz splits are counted as force-aligned timesteps at native sensor
rate.

| corpus | episodes | frames |
|---|---:|---:|
| pretrain (paired, both pairs) | 1,789 | 710,633 (64,603 optical + 646,030 array) |
| classification (paired, both pairs) | 599 | 239,184 (21,744 optical + 217,440 array) |
| force task (static + sliding modes) | 401 + 405 | 423,284 (static 209,869 + sliding 213,415) |
| slip task (sliding modes) | 405 | 213,415 |
| **total** | | **≈ 1.59M** |

Note: sliding episodes carry time-aligned 6D force as well, so they can also
serve force training; **all force results in the HTT paper use static mode
only** (the `force/` split), and slip results use the `slip/` split.

## Download

```bash
hf download AllenBi21/HTT-dataset --repo-type dataset --local-dir HTT-dataset
cd HTT-dataset && sha256sum -c MANIFEST.sha256   # optional integrity check
```

Dataloaders for every split ship with the HTT code release
([github.com/jxbi1010/HTT](https://github.com/jxbi1010/HTT)); pretrained
weights: [AllenBi21/HTT](https://huggingface.co/AllenBi21/HTT). Formats are
documented in [`FORMAT.md`](FORMAT.md).

## License

CC-BY-NC-4.0 (non-commercial research use).