Affordance Detection & Grasping

We have developed new methods for affordance detection & grasping. See our affordance work in IROS16, IROS17, ICRA18, IROS23, ICRA24, ICRA24. We have tackled language-driven grasping tasks with Grasp-Anything dataset and methods in CVPR24, ECCV24, IROS24.

Foundation & Generative Model

We have utilized foundation models for open-vocabulary tasks (IROS23, ICRA24) and developed VLM for robotic applications (Robotic-CLIP). We have proposed techniques for generative tasks such as scene synthesis (NeurIPS23) and dance generation (CVPR23, SIGGRAPH ASIA, ECCV24).

Efficient AI

Efficient models with real-time inference play an important role in robotics, medical, and manufacturing applications. We have developed lightweight models for autonomous driving (IROS20, IV22), medical imaging (TMI22), robotic grasping (Mechatronics23, IROS24) and federated learning (ICCV23).

Autonomous Catheterization

We have led research in developing world-first autonomous catheterization robot (ICRA20, ICRA20, TBME21, TBME22). We also developed CathSim, a realistic and high fidelity open-source simulator for endovascular intervention and the large-scale CathAction and Guide3D dataset.

Medical Imaging

We have proposed solutions to tackle challenging problems in medical imaging such as depth estimation (MICCAI21, MICCAI22), medical-VQA (MICCAI22), deformable registration (TMI22), sensing area detection (MICCAI23), and retina image (MICCAI24, MICCAI24).

Federated Learning

Collecting data to train big ML models violates the users’ privacy. We have proposed FL techniques to overcome this limitation. See our FL works for autonomous driving (IV22), non-IID data (ICRA24), topology design (ICCV23), and medical imaging (MICCAI23).