PhD Student · Robotics
I am a PhD student in Mechanical Engineering (Robotics) at the GRASP Lab, University of Pennsylvania, advised by Prof. Nadia Figueroa.
My research focuses on data-driven methods for adaptive and reactive robotic manipulation — enabling robots to infer task-relevant structure and adjust their control strategies in real time, toward safe and efficient deployment in complex, unstructured environments.
Prior to Penn, I received my B.S. in Aeronautical and Astronautical Engineering from Purdue University, where I worked on Bayesian transfer learning for composite manufacturing and led autonomous UAV development at the Air Force Research Lab.
We propose an adaptive impedance control framework combining online system identification with an interaction classifier, enabling robots to compensate for unknown end-effector loads while remaining passive to human perturbations and leveraging estimated contact forces as guidance for motion planning.
We propose an online, probabilistic framework for estimating the mismatch in the end-effector dynamics model, allowing impedance-controlled robots to manipulate heavy, unknown objects, showcasing fast and safe adaptation.
We introduce the concept of physics-guided transfer learning to predict the thermal conductivity of an additively manufactured short-fiber reinforced polymer (SFRP) using micro-structural characteristics extracted from tensile tests.
I implemented a conditional diffusion model for 2D point-mass navigation following the planning-as-denoising approach (Janner et al., ICML 2022). A U-Net learns to recover clean trajectories from Gaussian noise conditioned on start and goal positions, trained on D4RL Point Maze demonstrations, producing diverse collision-free paths at inference time.
Code(coming soon)
We implemented control barrier functions (CBFs) into an operational space impedance controller for a 7-DoF Franka Research 3 arm, enforcing joint limit, obstacle avoidance, and self-collision avoidance in real-time. The solver was implemented in Python with PyBullet and OSQP and evaluated on tasks requiring strict safety constraint enforcement.
We developed a full perception-to-execution pipeline for autonomous block stacking on a 7-DoF Franka Research 3, combining AprilTag-based detection, kinematic motion planning, and grasp execution. The system was evaluated on building stable towers under time constraints in both static and dynamic block configurations.
Led a 60-member interdisciplinary team as Project Manager and Lead Structures Engineer for NASA Student Launch, overseeing the full design-to-launch cycle of a high-powered competition rocket. Coordinated deliverables and communications directly with NASA and NAR, managed budget and timeline, and led structural design including FEA, CFD, and component manufacturing.
Implemented Neural Radiance Fields (NeRF) from scratch in PyTorch following the original formulation, including positional encoding, an MLP-based radiance field, and volumetric rendering via differentiable ray marching. Explored the effect of network capacity and sampling strategies on rendering quality across multiple views.
Implemented a GPT-inspired transformer from scratch in PyTorch following Andrej Karpathy's approach, including multi-head self-attention, positional encoding, and MLP blocks for character-level language generation on different datasets. Conducted ablation studies on regularization techniques including dropout, residual connections and layer normalization.