I am a Ph.D. student at GRASP Lab, University of Pennsylvania, advised by Prof. Nadia Figueroa. I want robot manipulation policies that are interactive and scalable. To get there, I have worked on data-efficient learning, learning beyond teleoperation, and force-aware manipulation. I am also passionate about designing innovative hardware.
Before coming to Penn, I received my MS in Robotics from Northwestern University and BS in Aerospace Engineering from UIUC.
May 2026 - Joining Dyna Robotics as a Research Intern.
May 2026 - VLMgineer is featured on Penn Today and Penn Engineering.
May 2026 - Invited talk on Flow with the Force Field at Nvidia Dex Team.
Feb 2026 - Flow with the Force Field is accepted to ICRA 2026.
Feb 2026 - VPP is accepted to ICRA 2026.
Jan 2026 - VLMgineer is accepted to ICLR 2026.
Sep 2025 - Joining Amazon Robotics as an intern working on Vulcan Stow.
Jun 2025 - EMP is accepted to IROS 2025.
Jun 2025 - OCR is accepted to IROS 2025.
April 2025 - Presented as a spotlight speaker at the ICRA 2025 Doctoral Consortium.
April 2025 - EMP is accepted as a spotlight paper @ Workshop on Structured Learning, ICRA 2025.
We present a pipeline that jointly leverages generated video and audio to recover both motion trajectories and contact force profiles from a single task description, enabling zero-shot force-aware manipulation where a kinematic-only baseline fails. Also presented at the Workshop on Manipulation Robustness: Towards Human-Level Robustness under Real-World Challenges, ICRA 2026.
We introduce a framework for generating force-informed data in simulation, instantiated by a single human demonstration, and show how coupling with a compliant policy improves the performance of a visuomotor policy learned from synthetic data. Also presented at the Workshop on Contact and Impact-aware Manipulation and the Workshop on Exploring the Role of Energy in Robot Learning and Control, IROS 2025.
We introduce VLMgineer, a novel VLM-driven evolutionary framework that automatically co-design tools and actions to solve robotics task. Also presented at the Workshop on Robot Hardware-Aware Intelligence, RSS 2025.
A quadratic programming-based control framework enforces viability theory constructed constraints on a passive controller tracking a dynamical system, ensuring the robot states remain within the safe set in an infinite time horizon. Best Paper Award at the Workshop on Exploring the Role of Energy in Robot Learning and Control and Best Student Paper Award at the Workshop on Building Safe Robots, IROS 2025.
A one-shot stable imitation learning framework that allows robots to adjust their behavior based on the scene change while respecting the task specification. Spotlight Paper at the Workshop on Structured Learning for Efficient, Reliable, and Transparent Robots, ICRA 2025.
We introduce MORF, a magnetic origami system that folds sheets into rigid, repeatedly reprogrammable structures with controllable fold angles, enabling adaptive robotic tools that can be reshaped for new tasks instead of rebuilt.
An object-centric recovery policy that detects when an imitation policy drifts out of distribution and steers it back via object keypoints, with no extra data. Spotlight Paper at the Workshop on Lifelong Learning for Home Robots, CoRL 2024.
A control architecture for torque-controlled robots that enforces safety constraints such as joint limits, self-collisions, external collisions, and singularities, while staying passive whenever the constraints allow.
A motion planning approach that learns plans with Neural Ordinary Differential Equations and certifies stability with Control Lyapunov Functions and safety with Control Barrier Functions, so learned motions stay stable and safe.
A dynamical-system representation for estimating human intent during dynamic object co-manipulation that accounts for the constraints of the human, letting the robot anticipate and follow its partner's motion.
A learning-from-demonstration method that encodes motion policies as elastic dynamical systems, so a learned policy adapts to new environment constraints without new demonstrations while preserving stability guarantees.
A novel statistical model that applies the Riemannian metric on the n-sphere to efficiently blend non-Euclidean directional data with Euclidean states for learning stable, time-independent motion policies
Using maximum entropy deep IRL to learn agent preference in continuous environment path planning
Developed software and hardware pipeline to play yoyo with visual feedback control on the Sawyer Robot arm
Performed receding horizon control using data-driven approach with small data in a continuous space