Dr Dandan Zhang
ROBOTICS · AI · EDUCATION
IMPERIAL COLLEGE LONDON
Perception
What can robots infer from touch and microscopic images?
We connect sensor-level computation with spatial representations, shared datasets and repeatable evaluation. Six selected outcomes span tactile perception and optical microrobot vision.
Research overview · All publications
01 / SENSING & REPRESENTATION
PixelTac · Perception inside the sensor
Communications Engineering · 2026 · Journal article
Outcome. A computational pixel array and microcontroller process tactile signals locally. On-chip force inference and mechanoreceptor-inspired responses reduce the need to send raw tactile images to an external computer.
ViTacTip · Vision, proximity and touch
IEEE Transactions on Robotics · 2025 · Journal article
Outcome. A transparent skin and biomimetic internal tips combine visual context with contact sensing. Benchmarking covers hardness, material and texture recognition; GAN-based interpretation separates information across sensing modalities.
SARL · Spatially aware visual–tactile learning
ICRA · 2026 · Conference paper
Outcome. Self-supervised objectives preserve local geometry in fused visual–tactile features. Evaluation across six downstream tasks shows stronger spatial representations, including improved edge-pose estimation compared with the tested self-supervised baselines.
02 / DATASETS & REPRODUCIBLE EVALUATION
TacVerse · Comparing and transferring across sensors
Advanced Robotics Research · 2026 · Journal article
Outcome. A dataset of 106,800 tactile images from seven sensors supports shape classification, grating classification and force regression. Within-sensor, cross-sensor and few-shot tests expose where transfer works and where sensor differences remain a barrier.
Paper · Project · Code · Dataset
TacEva · Task-specific sensor evaluation
Advanced Intelligent Systems · 2026 · Journal article
Outcome. Defined metrics and repeatable experimental pipelines compare tactile sensor performance across designs. The framework supports evidence-based sensor selection and identifies which aspects of a design need improvement for a particular task.
OTMR · Optical microrobot perception benchmark
MARSS · 2025 · Conference paper and dataset
Outcome. OTMR contains 232,881 microscope images spanning 18 microrobot types and 176 poses. Benchmarks of eight learning models compare pose classification and depth regression, providing a shared basis for microscopic perception research.
Original paper figures and detailed project presentations follow, including TacVerse, SARL, MultiDiffSense, TacEva, ViTacTip, MagicTac and Tac-VGNN.
 Add a TitleDescribe your image |  Add a TitleDescribe your image |  Add a TitleDescribe your image |
|---|---|---|
 Add a TitleDescribe your image |  Add a TitleDescribe your image |  Add a TitleDescribe your image |
ViTacTip
Design and Benchmarking
This paper presents ViTacTip, a compact multi-modal sensor for robotic perception. Its transparent see-through skin enhances vision and proximity sensing, while embedded biomimetic tips improve tactile and force perception. We validate its capabilities through multi-task learning for hardness, material, and texture recognition, together with benchmarking on object recognition, contact point detection, pose regression, and grating identification. A GAN-based framework further enables cross-modality interpretation and flexible switching between sensing modalities.
MagicTac
3D multi-layer design Based Tactile Sensor
This paper presents MagicTac, a high-resolution grid-based tactile sensor for robotic contact perception. Its 3D multi-layer design, inspired by the Magic Cube structure, improves spatial resolution, while multi-material additive manufacturing enables low-cost, repeatable, and assembly-friendly fabrication. Experiments on tactile reconstruction using deformation fields and optical flow show that MagicTac captures fine textures and dynamic contact information effectively. The sensor can be manufactured for as little as £4.76 in 24.6 minutes.
Design and Benchmarking of A Multi-Modality Sensor for
Robotic Manipulation with GAN-Based Cross-Modality Interpretation
Accepted by IEEE Transactions on Robotics
Poster Presentation at 2024 ICRA
MagicTac: A Novel High-Resolution 3D Multi-layer Grid-Based Tactile Sensor
Wen Fan, Haoran Li, Dandan Zhang
Paper Link:
https://ieeexplore.ieee.org/document/10610615
Accepted by 2024 ICRA
Tac-VGNN: A Voronoi Graph Neural Network for Pose-Based Tactile Servoing
Wen Fan, Max Yang, Yifan Xing, Nathan F. Lepora, Dandan Zhang*
Paper Link: https://arxiv-export3.library.cornell.edu/abs/2303.02708
Accepted by 2023 ICRA
Tactile pose estimation and tactile servoing are fundamental capabilities of robot touch. Reliable and precise pose estimation can be provided by applying deep learning models to high-resolution optical tactile sensors. Given the recent successes of Graph Neural Network (GNN) and the effectiveness of Voronoi features, we developed a Tactile Voronoi Graph Neural Network (Tac-VGNN) to achieve reliable pose-based tactile servoing relying on a biomimetic optical tactile sensor (TacTip). The GNN is well suited to modeling the distribution relationship between shear motions of the tactile markers, while the Voronoi diagram supplements this with area-based tactile features related to contact depth. The experiment results showed that the Tac-VGNN model can help enhance data interpretability during graph generation and model training efficiency significantly than CNN-based methods. It also improved pose estimation accuracy along vertical depth by 28.57% over vanilla GNN without Voronoi features and achieved better performance on the real surface following tasks with smoother robot control trajectories.
For more project details, please view our website: https://sites.google.com/view/tac-vgnn/home
