(ICLR 2026) AnyTouch 2: General Optical Tactile Representation Learning For Dynamic Tactile Perception

🚀 Welcome to the repo of AnyTouch 2! If our project helps you, please give us a star ⭐ on GitHub to support us. 🙏

📑To Do

• Quick Start Demo Code
• Dataset Pre-processing
• Sparsh Evaluation Code

🛠️ Requirements and Installation

1. Create Environment

   conda create -n anytouch2 python=3.9
   conda activate anytouch2
   

2. Install PyTorch 2.4.0 + Cuda 12.4

   pip install torch==2.4.0 torchvision==0.19.0 --index-url https://download.pytorch.org/whl/cu124
   

3. Install other required packages:

   git clone https://github.com/GeWu-Lab/AnyTouch2.git
   cd AnyTouch2
   pip install -r requirements.txt
   

🚀 Quick Start

1. Download AnyTouch 2 Model Checkpoints into checkpoints/ (Complete the form to get access first)

   huggingface-cli download --repo-type model xxuan01/AnyTouch2-Model --local-dir checkpoints
   

   Checkpoint Performance:

   		TAG	Cloth	Slip / Delta Force (Sparsh)	Force (Sparsh)	Force (ToucHD)
   num_frames	stride	Acc ↑	Acc ↑	F1 Score ↑ / RMSE ↓	RMSE ↓	RMSE ↓
   4	2	76.97	42.31	86.66 / 87.80 (DG) 97.96 / 80.83 (Mini)	624.26 (DG) 202.14 (Mini)	894.32 (DG) 1051.03 (Mini)
   2	6	74.15	40.76	86.60 / 83.15 (DG) 97.85 / 89.21 (Mini)	643.91 (DG) 208.41 (Mini)	1076.33 (DG) 1311.27 (Mini)

2. Run quick_start.sh (Coming Soon)

   bash scripts/quick_start.sh
   

🤖 Downstream Evaluation

1. Download ToucHD (Force) (Complete the form to get access first), Touch and Go an Cloth into datasets/

   ### Download ToucHD (Force). Please complete the form to get access first.
   huggingface-cli download --repo-type dataset xxuan01/BAAI/ToucHD-Force --local-dir datasets
   

2. Pre-process the datasets (Coming Soon)

3. Run scripts to start downstream training and evaluation

   bash scripts/run_probe_tag.sh
   bash scripts/run_probe_cloth.sh
   bash scripts/run_probe_touchd.sh
   

Coming Soon!