Ph.D., Electrical Engineering

Master of Science, Mechanical Engineering

Bachelor of Engineering, Mechanical Engineering

Fauna Robotics, New York, NY

New York University, Tandon School of Engineering, Brooklyn, NY

• Proposed a new methodology for synthesizing control barrier functions from raw point cloud measurements.
• Utilized GPU acceleration to make the control barrier function synthesis computationally efficient.
• Performed experimental validation of the proposed approach on Unitree Go2 and Unitree B1 quadrupedal robots.

New York University, Tandon School of Engineering, Brooklyn, NY

• Proposed a new methodology for control barrier function synthesis using differentiable optimization.
• Performed experimental validation of the proposed methodology on the Franka Research 3 robot.
• Extended the methodology to handle problems with time-varying safe sets while considering measurement noise and actuation limits.

New York University, Tandon School of Engineering, Brooklyn, NY

• Proposed a new methodology for learning-based control barrier function synthesis that starts from handcrafted control barrier functions.
• Proposed a prioritized sampling method to make learning-based control barrier function synthesis more data-efficient.

New York University, Tandon School of Engineering, Brooklyn, NY

• Proposed a new methodology for state-constrained nonlinear stochastic optimal control using forward-backward stochastic differential equations and LSTMs.
• Created custom simulation environments to test the performance and scalability of nonlinear systems with both continuous and hybrid dynamics.

SafeAI Inc, San Jose, CA

• Created a reinforcement learning simulation environment for the load-haul-dump cycle.
• Designed the reward function, state space, and action space to be realistic while also accelerating training.
• Constructed a behavior tree that orchestrates reinforcement-learning-based and traditional controllers.

Carnegie Mellon University, Robotics Institute, Pittsburgh, PA

• Proposed a 6-degree-of-freedom (DOF) joint-based kinematic model for a multi-link bipedal robot system.
• Developed a 6-DOF joint-based kinematic identification algorithm using linear regression and achieved 92.3\% accuracy in simulation with white-noise-polluted data.
• Implemented the kinematic identification algorithm on a real bipedal robot ATRIAS using mocap data.

Huazhong University of Science and Technology, Mechanical Engineering Department, Wuhan, Hubei, China

• Used a three-dimensional curvature-based model to represent the whole femur-knee-tibia system, which overcame the difficulty of modeling non-uniformly shaped contact parts in bio-joints.
• Implemented a modal-superposition method to reduce the number of sensors required to only three.