Portfolio item number 1
Short description of portfolio item number 1
Posts by Collection
portfolio
Portfolio item number 2
Short description of portfolio item number 2
publications
Learning to Adapt in Dynamic, Real-World Environments Through Meta-Reinforcement Learning
Published in International Conference on Learning Representations (ICLR), 2018, 2018
We apply meta-reinforcement learning to enable fast adaptation in dynamic real-world environments, demonstrated on a millirobot that adapts to leg loss, terrain changes, and payload variation using 1.5 hours of offline data.
On Learned Sketches for Randomized Numerical Linear Algebra
Published in arXiv preprint, 2020, 2020
We study learning-based approaches to sketch design for randomized numerical linear algebra, characterizing when learned sketches outperform random ones.
Safe Control Under Input Limits with Neural Control Barrier Functions
Published in Conference on Robot Learning (CoRL), 2022, 2022
We scale safety filter synthesis to high-dimensional systems by training a neural control barrier function, synthesizing a filter for a 10D quadrotor–pendulum system that triggers orders of magnitude less often than MPC.
Learning the Positions in CountSketch
Published in International Conference on Learning Representations (ICLR), 2023, 2023
We show that the hash positions in CountSketch can be learned from data to improve performance, and provide theoretical guarantees. Oral presentation (top 25%) at ICLR 2023.
Safety Index Synthesis via Sum-of-Squares Programming
Published in American Controls Conference (ACC), 2023, 2023
We formulate safety index synthesis as a sum-of-squares programming problem, enabling automated construction of control barrier functions for nonlinear systems with formal safety guarantees.
Synthesis and Verification of Robust-Adaptive Safe Controllers
Published in European Controls Conference (ECC), 2024, 2024
We consider safety filter synthesis for systems with uncertain model parameters, devising a sum-of-squares programming algorithm that generates a geofencing filter for a quadrotor with unknown drag in minutes.
Certifying Robustness of Learning-Based Keypoint Detection and Pose Estimation Methods
Published in ACM Transactions on Cyber-Physical Systems, 2025, 2025
We formally verify the robustness of learned perception components (keypoint detection, pose estimation) in an aircraft autonomy stack, providing certificates against adversarial perturbations.
Approximately Optimal Global Planning for Contact-Rich SE(2) Manipulation on a Graph of Reachable Sets
Published in IEEE Transactions on Robotics (under review), 2026
We devise a hierarchical planner that integrates contact-aware trajectory optimization, collision-free motion planning, and MIP-based graph search over reachable set primitives, achieving >60% lower cost plans than leading methods for bimanual contact-rich manipulation.
talks
Sample Inefficiency in Deep Reinforcement Learning for Robotics
Published:
Workshop talk examining sample efficiency challenges in applying deep reinforcement learning to real-world robotics problems, motivated by our work on model-based RL for adaptive locomotion.
Scalable Synthesis of Safe Control Filters
Published:
Invited talk at the Safe Autonomous Systems Lab at UCSD on scalable safety filter synthesis using neural control barrier functions.
Scalable Synthesis of Safe Control Filters
Published:
Conference presentation of our work on neural control barrier functions for safe control synthesis in high-dimensional systems (10D+).
Synthesis of Safe Control Filters for Uncertain Systems
Published:
Conference presentation of our paper on synthesis and verification of robust-adaptive safe controllers, including a sum-of-squares programming approach for safety filter synthesis under parametric uncertainty.
Reactive Safety for Self-Driving
Published:
Awardee talk presenting research on reactive safety filters for self-driving systems, including synthesis methods for uncertain and high-dimensional systems.
High-Performance Planning for Contact-Rich Manipulation
Published:
Invited talk at the CMU Manipulation Seminar on hierarchical global planning for contact-rich manipulation using reachable set primitives.
teaching
Efficient Algorithms and Intractable Problems (CS 170)
Undergraduate course, UC Berkeley, 2017
Teaching assistant for UC Berkeley CS170: Efficient Algorithms and Intractable Problems (Fall 2017 and Fall 2018), taught by Profs. Umesh Vazirani, Prasad Raghavendra, and Sanjam Garg. A core undergraduate algorithms course covering divide-and-conquer, dynamic programming, graph algorithms, NP-completeness, and approximation algorithms. Recognized among the top 10% of TAs in the UC Berkeley EECS Department.
Introduction to Artificial Intelligence (CS 188)
Undergraduate course, UC Berkeley, 2019
Teaching assistant for UC Berkeley CS188: Introduction to Artificial Intelligence, taught by Prof. Sergey Levine. A large undergraduate course covering search, constraint satisfaction, Bayesian networks, machine learning, and reinforcement learning.
Math Fundamentals for Robotics (16-811)
Graduate course, Carnegie Mellon University, 2021
Teaching assistant for CMU 16-811: Math Fundamentals for Robotics, taught by Prof. Michael Erdmann. A graduate-level course covering the mathematical foundations of robotics including linear algebra, calculus, probability, and optimization.
Kinematics, Dynamics, and Control (16-711)
Graduate course, Carnegie Mellon University, 2022
Teaching assistant for CMU 16-711: Kinematics, Dynamics, and Control, taught by Prof. Hartmut Geyer. A graduate-level course covering rigid body kinematics, dynamics, and control theory with applications to robotics.