About Me

COBRA is a 2m long snake robot that operates untethered, with a stereocamera and processor in the head for localization, mapping and control. It was initially developed as a proof of concept for the NASA BIG Idea Competition in 2022, which called for energy-efficient alternate mobility solutions for craters on the lunar south pole. The original team was 13 undergraduate students from various backgrounds, with me as the sole graduate student in an advisory capacity; together we showed basic manual control with simple gaits. After winning the competition, COBRA became the focus of my PhD work, where I led the research to expand the gait library, integrate perception, and run SLAM onboard. I also developed a contact-implicit model predicting the robot’s motion under various gaits, and demonstrated autonomous behaviors such as path tracking and loco-manipulation.

M4 is a wheeled robot that can morph into a UAV to fly. Designed at Caltech by my PhD advisor with a post-doctoral student, the first version showed the transforming behavior and various locomotion modes, combining thrusters and wheels to traverse a wide variety of terrain. I became involved with later versions, writing control firmware and contributing to experimental demonstration of its capabilities. I guided a few MS students through thesis research into traversability estimation for multi-modal path planning, to autonomously decide when to drive or fly. I also explored morphing in flight for fault-tolerant, agile flight, and led the development of a new 3D-printed lightweight version at Northeastern, later used with a humanoid robot in a multi-robot collaboration demo at Caltech.

Aerobat is a small 30g bat-robot with a payload capacity of just 15g for all sensing and compute. Flapping-wing robots suit confined spaces where collisions with the walls are likely, contact-rich locomotion like wheels or legs is not possible, and rotary-wing platforms struggle with turbulence from their own wake. Open questions remain on their control, and on state estimation for autonomous operation. I first worked on Aerobat for my MS thesis, integrating a lightweight sensing and compute package for visual-inertial odometry. Since moving to COBRA for my PhD, I have continued collaborating with the team, building dynamic models and state estimation.

Husky is a lightweight quadruped with propellers on its back for aerial and thruster-assisted ground locomotion. This extends its range beyond standard quadrupeds — walking on narrow paths, steep slopes, and low-friction surfaces, or flying over terrain that is inaccessible or unsuitable for legs. Through my PhD I have collaborated on this platform, building firmware, models, and experimental demonstrations of the multi-modal operation.

During my Master’s, I did a co-op (internship) at the Woods Hole Oceanographic Institution (WHOI), working with Amy Kukulya on REMUS and LRAUV robots. The LRAUV is designed for long-range missions, controlled remotely over a cellular network by operators on the shore. I was involved with field operations and software support as we collected ocean data off the coast of Cape Cod, Massachusetts.

The experience reshaped how I approach robot operation and software. Because pressurized hulls seal the electronics during a mission, onboard systems are largely inaccessible once deployed, with the only access being through low-bandwidth cellular or Wi-Fi. To ensure the success of long missions where robot recovery is not straightforward, everything was built with layered failsafes, fault-sensing and diagnostic tools, and options for remote fixes. This was enabled by designing the architecture from the ground up with modularity in mind, avoiding deep dependency trees that would be hard to maintain on the fly. I’ve carried many of the practices I learned there into my work since.