cv

Graduate Researcher

DAIR Lab and GRASP Lab, University of Pennsylvania

I work at DAIR Lab under my supervisor Michael Posa. My research focuses on model learning and real-time control for robotic contact-rich tasks.

• Contact-Implicit MPC
• Contact-Rich Modeling
• Model Learning

Robotics Engineer (Founding Engineer)

Eureka Robotics, Singapore

I worked at Eureka Robotics as a Robotics Engineer. I was responsible for the development of the Archimedes and Pytagoras Robots, both are ultra-precise optics handling robotic systems.

• Archimedes Robot
• Pytagoras Robot

PhD Student in Robotics

University of Pennsylvania, Philadelphia, USA

Robotics

• Contact-Implicit MPC
• Contact-Rich Modeling
• Model Learning

Bachelor of Mechanical Engineering in Robotics and Mechatronics (Highest Distinction)

Nanyang Technological University, Singapore

Robotics

• Robotics
• Mechatronics
• Control Systems

Graduate Research Fellowship

Mechanical Engineering and Applied Mechanics (MEAM) Department at University of Pennsylvania

Active Tactile Exploration for Rigid Body Pose and Shape Estimation

Under Review

This work presents a tactile-only learning and exploration framework for estimating the shape and location of previously unseen rigid objects. By formulating a physically consistent, contact-rich loss and using information-driven exploration, the approach learns accurate object geometries with minimal motion and under 10 seconds of data, achieving efficient and robust performance in both simulation and real-world experiments.

Push Anything: Single- and Multi-Object Pushing From First Sight with Contact-Implicit MPC

Under Review

This work demonstrates that contact-implicit MPC can scale beyond curated examples to complex, multi-object planar pushing. By introducing the faster C3+ algorithm and a full perception-to-hardware pipeline, the approach enables real-time, contact-rich manipulation across diverse object geometries, achieving high success rates and precise goal attainment even in challenging multi-object scenarios.

Enhancing Task Performance of Learned Simplified Models via Reinforcement Learning

ICRA 2024

This work proposes learning simplified dynamics models for contact-rich manipulation by directly optimizing task performance rather than prediction accuracy. By combining policy gradient methods with MPC and applying them to a three-fingered hand manipulating unknown objects, the approach improves success rates by up to 15% while remaining data-efficient, achieving strong performance with under 30 minutes of data.

Reinforcement Learning for Reduced-order Models of Legged Robots

ICRA 2024

This work bridges model-based control and model-free reinforcement learning for bipedal locomotion by learning a reduced-order model within an MPC framework, combining physical interpretability with adaptability to new task commands. The resulting approach improves performance, achieving a 49% expansion of the viable task region and a 21% reduction in motor torque cost compared to prior methods.

An ultralightweight and living legged robot

Soft Robotics

This study presents an ultralight, low-power legged robot based on a living beetle equipped with a wireless stimulator, enabling versatile locomotion and turning through simple antenna stimulation. Leveraging the insect’s natural compliance and adaptability, the system achieves robust terrain traversal with minimal control complexity and power consumption in the microwatt range.