Research

My research develops tactile intelligence for robots, spanning high-resolution tactile sensing, physically grounded tactile simulation, contact-aware manipulation, and multimodal foundation models. Publications are organized below by research direction.

Visuotactile Sensing

Develop high-resolution, high-precision visuotactile sensors for robotic fingertips; develop multidimensional, high-resolution wearable tactile fingertip sensors for capturing human manipulation skills and transferring them to robots; and develop soft six-axis force/torque sensors with tool-side pose sensing as versatile interfaces for robotic systems, such as wrist-mounted force sensing.

GelStereo 2.0: An Improved GelStereo Sensor With Multimedium Refractive Stereo Calibration

Chaofan Zhang, Shaowei Cui, Shuo Wang, Jingyi Hu, Yinghao Cai, Rui Wang, and Yu Wang

IEEE Transactions on Industrial Electronics, 71(7), 7452–7462, 2024

Paper

Keypoints

• Compact stereo visuotactile sensor for high-precision 3D contact reconstruction.
• Refractive stereo ray tracing (RSRT) model for physics-based modeling of light propagation through multiple media within the sensor.
• Efficient multi-depth calibration improves reconstruction accuracy across the sensing surface.

Learning-Based Six-Axis Force/Torque Estimation Using GelStereo Fingertip Visuotactile Sensing

Chaofan Zhang, Shaowei Cui, Yinghao Cai, Jingyi Hu, Rui Wang, and Shuo Wang

IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 3651–3658, 2022

Paper

Keypoints

• Jointly encodes 2D marker motion and 3D tactile deformation for six-axis force/torque sensing.
• Contact positional encoding compensates for location-dependent force characteristics.
• Demonstrates real-time force-feedback adaptive grasping with GelStereo fingertips.

SpikingTac: A Miniaturized Neuromorphic Visuotactile Sensor for High-Precision Dynamic Tactile Imprint Tracking

Tianyu Jiang, Chaofan Zhang, Shaolin Zhang, Shaowei Cui, and Shuo Wang

IEEE Transactions on Industrial Electronics, 2026

Paper

Keypoints

• Miniaturized standalone event-camera module enables compact neuromorphic tactile sensing.
• Global dynamic state mapping and denoising support high-rate tactile imprint tracking.
• Hysteresis-aware incremental updates improve zero-point stability and dynamic accuracy.

Tactile Simulation

Physics-based simulation of elastomer deformation and optical tactile rendering across sensor geometries, coatings, and output modalities.

TacFlex: Multimode Tactile Imprints Simulation for Visuotactile Sensors With Coating Patterns

Chaofan Zhang, Shaowei Cui, Jingyi Hu, Tianyu Jiang, Tiandong Zhang, Rui Wang, and Shuo Wang

IEEE Transactions on Robotics, 41, 3965–3985, 2025

Paper Website Code

Keypoints

• Couples finite-element deformation with optical rendering for physically grounded tactile simulation.
• Support multi-type visuotactile sensors and integrate multi-mode tactile imprints.
• Generates multi-mode tactile imprints for scalable Sim2Real perception and manipulation learning.

Dexterous Manipulation with Vision and Touch

Develop visual-tactile fusion and skill learning methods to enhance the robot's precise and dexterous manipulation capabilities through tactile perception.

DexTac: Learning Contact-Aware Visuotactile Policies via Hand-by-Hand Teaching

Xingyu Zhang, Chaofan Zhang, Boyue Zhang, Zhinan Peng, Shaowei Cui, and Shuo Wang

IEEE Transactions on Automation Science and Engineering, 23, 11171–11182, 2026

Paper

Keypoints

• Hand-by-hand teaching captures motion trajectories, force distributions, and spatial contact regions.
• Contact-aware policy learning maintains effective contact throughout complex interactions.
• Demonstrates robust unimanual injection, including high-precision tasks with small syringes.

Sim-to-Real Assembly Learning Using Rich Visual–Tactile Perception

Chaofan Zhang, Shaowei Cui, Xiaoge Cao, and Shuo Wang

IEEE Transactions on Industrial Informatics, 2026

Paper

Keypoints

• Oracle-to-embodiment framework transfers privileged simulation knowledge to real-world sensory observations.
• Task-aware sensory gating adaptively fuses visual and tactile information across different assembly phases.
• Enables data-efficient Sim2Real peg-in-hole insertion under tight clearances and unseen conditions.

Tactile Foundation Models

Incorporate tactile modality into multimodal foundation models for robotic manipulation, enabling more generalizable and precise manipulation.

VTLA: Vision-Tactile-Language-Action Model with Preference Learning for Insertion Manipulation

Chaofan Zhang, Peng Hao, Xiaoge Cao, Xiaoshuai Hao, Shaowei Cui, and Shuo Wang

Biomimetic Intelligence and Robotics, 100333, 2026

Paper Website

Keypoints

• A Vision–Tactile–Language–Action model for insertion manipulation.
• Preference learning enhances continuous control and insertion performance.
• Uses exclusively TacFlex-generated simulation data for training and achieves zero-shot Sim2Real transfer.

FG-CLTP: Fine-Grained Contrastive Language Tactile Pretraining for Robotic Manipulation

Wenxuan Ma, Chaofan Zhang, Yinghao Cai, Guocai Yao, Shaowei Cui, and Shuo Wang

IEEE Robotics and Automation Letters, 11(9), 2026

Paper

Keypoints

• Aligns tactile signals with language descriptions of fine-grained 3D contact states rather than coarse semantic labels.
• Discretized numerical tokens preserve measurable contact geometry and force-related structure.
• Supports sensor-general tactile representations and tactile-language-action policy learning.

FeelWorld: Visuo-Tactile World Model for Hierarchical Contact Prediction and Planning

Wenxuan Ma, Chaofan Zhang, Chao Xue, Yinghao Cai, Guocai Yao, Shaowei Cui, and Shuo Wang

arXiv preprint arXiv:2607.24267, 2026

Paper

Keypoints

• Jointly predicts visual latents and hierarchical tactile states: contact, 3D tactile state, and slip.
• Contact-gated asymmetric attention preserves visual prediction before contact and activates tactile fusion when needed.
• Predicted contact and slip states support contact-aware CEM planning for grasping and insertion.