Senior Robotics Engineer · Boston, MA
I build the software that makes industrial robots move: motion planning, manipulation, control, and the interfaces that let the outside world drive them. Currently building Locus Array, a new class of fully autonomous fulfillment robots, at Locus Robotics.
I'm a robotics software engineer focused on kinematics, motion planning, control, and the real-time interfaces that expose a robot to the world.
I've worked across industrial robotics at Kawasaki, KUKA, and now Locus Robotics, building everything from core robot OS modules to the tooling and applications that ship on top of them.
My favorite problems are in motion: solvers that respect torque and acceleration limits while staying smooth, planners that thread constrained spaces, and control interfaces that make the robot easy to drive.
Outside of work I prototype constantly: motion planning playgrounds, drones, VLA test benches. I stay deep in the open-source stack: ROS 2, Gazebo, Drake, and whatever simulator is fastest this year.
Senior Robotics Engineer Current
I develop the software that runs the Locus Array robot, Locus's new Robots-to-Goods platform: a fully autonomous fulfillment robot that drives to inventory and picks, puts away, and replenishes on its own.
A machine like that is a lot of software working in concert, and my work runs through all of it, wherever the robot needs it that week. Beyond that I can't say much: most of what makes Array tick stays in the building.
Senior Robotics Software Engineer
Developed core robot OS modules for motion and external control, and designed external control interfaces focused on ease of use, comprehensive functionality, and scalable deployment. Co-developed an inverse kinematics and motion library built on saturation-in-the-nullspace (SNS) algorithms.
Flagship
An external jogging interface that pairs a 3D space mouse with the SNS solver: twist input becomes end-effector velocity through velocity IK, integrated into joint position commands streamed over ROS 2 control and the RTC driver. The solver enforces torque and acceleration limits internally, so jogging stays smooth even on high-payload arms.
Constraint modes lock individual linear or rotational axes, and a relaxed-axis mode frees a chosen DOF. Relax tool RY, for example, and the solver uses the resulting nullspace to pass cleanly through singularities and joint limits.
Robotics Software Engineer
Built applications on Kawasaki Real-Time Control (RTC) and the Robot Network API, and consulted for universities, startups, and enterprises adopting RTC.
Shipped internal tooling in C++, Python, and C#: automated robot configuration, purchase-order tracking, data logging and visualization, QA test suites, and a robot programming IDE. Also built 2D/3D vision demo systems.
Flagship
A live body-tracking follow/teach application: Azure Kinect skeleton tracking drives the arm to mirror an operator's movements in real time, recording positions and motions the way a teach pendant would. Built on RTC and ROS with MoveIt and the Azure Kinect ROS driver.
Team entry to Intrinsic's AI for Industry Challenge: a fine-tuned SmolVLA vision-language-action policy teaching a simulated UR5e to insert fiber-optic connectors, scoring 110.3 against the provided baseline of 38.2, nearly 3 times better. Built with Abhijit Makhal and Amrit Saini; full write-up inside.
Meta's V-JEPA 2-AC video world model picking and placing in MuJoCo and Isaac Sim with no task training. It plans by imagining moves and matching goal photos. Pick-and-place went from 0 of 6 to 28 of 40; how the model works and what broke, inside.
Graduate SLAM course project at Carnegie Mellon evaluating ORB-SLAM2's robustness in urban driving scenes. Benchmarked the stock pipeline on the KITTI dataset against degraded input streams to quantify sensitivity to motion blur and other environmental interference.
Senior design project: a VR game you control with your brain. EEG from an OpenBCI headset streams over LSL into Unity, where a focus heuristic built on delta and theta band power drives in-game magic.
SVMs, decision trees, and DNNs were evaluated for classification.
A complete Zelda-inspired adventure game built in Unity with a four-person scrum team: full-length story, dungeons, and bosses.
I owned the backend systems: inventory, items, abilities, shops, game state, and UI. The game was selected for the university showcase.
An idle/tycoon game built in Godot 4 as a deliberate agentic software engineering case study: taking a project from an empty editor to a playable build through an AI-assisted workflow.
Each kitchen is a procedurally generated floor plan, built with a Voronoi tessellation and Lloyd relaxation chosen from a sandbox that benchmarked at least 13 generation algorithms.
A from-scratch playground for motion planning algorithms, visualized in Python and Pygame. Hosts PRM construction with shortest-path queries, with a roadmap toward A*, Hybrid A*, and multi-DOF robots executing planned paths in simulation.
The professional record is on LinkedIn; the code is on GitHub.