Innovation. Turns intrinsic tactile sensor noise into self-calibrated control thresholds, removing the need for external force calibration, material models, or manual threshold tuning. Three coordinated reflexes address slip, excessive grip, and overload for force-sensitive manipulation.
arXiv ↗
Innovation. Uses tactile feedback as a teacher during data collection: a high-rate controller shapes demonstrations so ACT and π0.5 can learn safer grasping without tactile inputs at inference. The work connects demonstration quality with learned contact behavior and distinguishes what policies can learn from what still requires real-time feedback.
Lightweight visual recognition for embedded edge-AI applications. My contributions include edge-computing algorithm design, deployment, and acceleration.
My research lies at the intersection of robot learning, multimodal perception, and physical interaction. I am interested in how vision-language-action models, tactile sensing, and reinforcement learning can help robots understand tasks and manipulate objects reliably in the real world.
I am particularly interested in contact-rich and force-sensitive manipulation, the role of sensory feedback in learning and control, and generalization across objects and environments. My work combines learning-based methods with closed-loop control, with an emphasis on real-robot validation and robust physical behavior.
Research on tactile sensing and robot learning for real-world manipulation.
Research on tactile simulation and robot learning.
Developed a multi-task robot manipulation system using reinforcement learning.
Desktop Robot Track champion, Embodied AI Program.
MPhil in Robotics and Autonomous Systems
B.Eng. in Automation
First-Class Scholarship · Outstanding Graduate · Innovation Award
