Research

Dr. Tao's research focuses on learning-based robot control and embodied intelligence, with the goal of enabling robots to perform dexterous manipulation autonomously and to collaborate with human operators naturally, intuitively, and safely. His work spans the full stack from tactile sensing hardware to control policies and human-robot interfaces, and is organized around four thrusts:

1. Dexterous Telemanipulation and Bidirectional Haptic Interaction. Developing glove-free, end-effector-oriented control paradigms that decouple human command from robot embodiment, allowing operators to command high-DoF robotic hands without wearable trackers or kinematic retargeting. This includes the Bi-Hap system for momentum-based haptic feedback, learning-based motion mapping and assistance strategies that adapt to operator preference, and diffusion-based anomaly recovery for robust vision-based tracking during teleoperation.

2. Multi-Modal Tactile Sensing and Robotic Perception. Designing cost-effective, manufacturable tactile sensing hardware and the perception models that use it. The Bio-Skin sensor provides normal force, shear, and temperature sensing with active thermoregulation at low unit cost, and related work addresses data-driven optimization of sensor placement, cross-sensor representation learning, and visuo-tactile foundation models for grasping.

3. Learning-Based Dexterous Manipulation and Sim-to-Real Transfer. Advancing the learning capability of autonomous manipulation through adaptive hierarchical curricula, multi-agent formulations of finger cooperation, physics-guided reward design, and curriculum-based sensing reduction that narrows the simulation-to-reality gap. Recent work extends to scalable data generation in simulation and VR, and to explicit diffusion and multi-modal policies for generalized multi-task manipulation.

4. Safe and Secure Human-Robot Systems. Investigating learning strategies for human-robot cooperation under general or partially specified goals, risk-aware interaction through multi-modal simulation and physics-informed models, and control-theoretic safety guarantees for redundant manipulators. A complementary line examines the cybersecurity and privacy vulnerabilities of robotic sensing, including adversarial attacks on mmWave perception.