Working on robotics tools that help others in the open source robotics community brings me an unreasonable amount of inexplicable joy. I recently released open source ROS2 drivers for the second generation Vive / Valve lighthouse system, which are based on the libsurvive project. I am currently working on integrating the tracy frame profiling tool with ROS2, so that others can get real-time performance monitoring for their nodes. 🔎

In the long term, however, my goal is to work on a project that lowers the barrier-to-entry for others when it comes to working on hybrid perception models -- those that combine traditional approaches to estimation (physics, filtering, covariance, etc) with data-driven / learnable elements. Motivated by the recent work in differentiable filters and factor graphs, I'd like to write ROS2 tooling to help make it easier to build, train and deploy these types of models to both arm64 and amd64 targets -- sort of like if GTSAM and jax give birth to a small child running on a Google Coral AI Board. 🧠

My past projects include helped write ROS code for NASA's Astrobee robot -- specifically on the motion planning subsystem. You can watch a presentation about that here. I also worked for a couple of years in the self-driving car industry, where I learned a lot about robotics tooling, data pipelines, containerization and test-driven architectures. Right now I'm back at NASA working for the most part on a lunar rover, but I also spend some of my time helping port a ROS2-based cooperative self-driving stack to a Jetson architecture, so that it can be run at 1/10th scale on little radio control cars. 🚗