TacForeSight: Force-Guided Tactile World Model for Contact-Rich Manipulation

Yujie Zang^*,1,2,3, Yuhang Zheng^*,2, Xian Nie^*,1,3, Yupeng Zheng⁴, Shuai Tian⁴, Songen Gu⁵, Chen Gao², Zining Wang¹, Shuicheng Yan^✉,2, Wenchao Ding^✉,1

¹ TARS Robotics ² National University of Singapore ³ Shanghai Jiao Tong University ⁴ Institute of Automation, Chinese Academy of Sciences ⁵ Fudan University

^* Equal contribution ^✉ Corresponding author

Paper Code Dataset Checkpoint

Overview

Method

Overview of TacForeSight. Our framework consists of two coupled components. In Stage 1, a force-conditioned tactile world model encodes dual-finger tactile fields into compact latent representations and predicts tactile evolution conditioned on wrist force/torque signals. In Stage 2, the predicted tactile dynamics are used as contact priors for a lightweight flow-based policy.

Demo

We demonstrate real-time servo-control deployment for fast contact regulation, 4× position-control evaluations for precise manipulation, and tactile visualization for interpreting predicted contact evolution.

Real-Time Deployment 1× speed

Card Swiping

Vase Wiping

In-Process Perturbations Tasks 4× speed

Height Perturbation during Wiping

Angle Perturbation during Swiping

Pose Perturbation during Tube Adjustment

Illumination Disturbance Tasks 4× speed

Vision-based Model

TacForeSight

Vision-based Model

TacForeSight

Contact-Rich Manipulation Tasks 4× speed

Vase Wiping

Card Swiping

Tube Adjustment and Insertion

Bulb Insertion and Locking

Wire Insertion

Tactile Generation

Tactile Foresight Visualization

Citation

If you find TacForeSight useful in your research, please cite:

@article{tacforesight2026,
  title   = {TacForeSight: Force-Guided Tactile World Model for Contact-Rich Manipulation},
  author  = {Yujie Zang and Yuhang Zheng and Xian Nie and Yupeng Zheng and Shuai Tian and Songen Gu and Chen Gao and Zining Wang and Shuicheng Yan and Wenchao Ding},
  journal = {TBD},
  year    = {2026}
}