My research leverages tools from optimization, geometry, and information theory to develop robots that are capable of exploring and understanding the world around them. I focus on applications in the Earth and environmental sciences, where advances in the science of robotics (the fundamental algorithms underpinning robot intelligence) will lead to breakthrough capabilities in the data we can gather on the science of our natural world. This means building robots that can robustly recognize where they are in remote and austere environments (e.g., underwater, many kilometers underneath arctic ice shelves), reconstruct what they have seen (e.g., how is temperature distributed throughout the water or what is the shape of this glacier), and make decisions on how to obtain the most informative next measurements.
Our lab approaches these largely through the lens of Bayesian inference for these localization and reconstruction challenges, where a robot’s position can be thought of as the “most likely location, given a set of noisy measurements” and the distribution of temperature or geometry of a glacier can be framed similarly. Then choosing the next best place to take a measurement is can be similarly framed as determining the next measurement(s) to add to this set to optimally reduce uncertainty in our estimate. As is often the case, the limiting factor is computational efficiency and reliability of the algorithms. This is where we apply tools from advanced optimization and geometry to develop efficient algorithms to solve these problems.
Please describe one or two of your most interesting projects.
My most exciting project focused on developing low-cost localization systems for underwater robots and (in pilot experiments) demonstrated we could reduce the cost of reliable navigation for these robots by over an order of magnitude (from ~$200k to ~$10k).
Right now, robots are the best way for us to measure some really valuable quantities in remote, yet critical parts of the world (e.g., in melting zones of glaciers). However, the navigational equipment alone for these robots is so expensive ($200k-500k) that there is little hope of scaling these measurements. I looked at whether we could use much lower-cost sensors (~10k) that would use acoustics to measure the distance from the robot to a fixed beacon or other robots, but the challenge here was that these low-cost sensors led to much more difficult inference problems within the localization algorithms. My breakthrough here was developing an algorithm that leveraged advanced tools from optimization (semidefinite programming and manifold optimization) to guarantee that we could always solve the inference problem (under well-understood sensor error regimes). This provided the reliability necessary to use these lower cost sensors. We tested these algorithms on a whole host of robotics problems, but the most relevant being a river environment. We showed that our $10k setup got the same navigational quality as expected of the $200k system shipped on current marine robots.
How did you end up where you are today? (Your research journey)
My research journey is a winding one marked by having many great mentors and many lucky opportunities. After I finished my undergrad I was excited about the idea of putting robots into the unexplored and was lucky to discover a PhD program between MIT and the Woods Hole Oceanographic Institution, one of the world’s preeminent ocean research institutions, that was the perfect fit for this type of work. I was lucky to be accepted and then spent my PhD with many great colleagues both advancing the state of the art in what robots can do and doing basic science to better understand the ocean and our climate systems. I wanted to do work that felt like it really could make a difference in the world, and this intersection of robotics and earth sciences seemed like exactly the opportunity to do that. With both inspiring colleagues and work that was the perfect outlet for what impact I wanted to make on the world, I’ve been increasingly pulled into this field of environmental and climate robotics.
What is the most significant scientific contribution you would like to make?
I want to develop robot observing systems that are so effective, efficient, and scalable that we can provide scientists studying the Earth’s natural systems (oceans, atmosphere, and land) with all of the data necessary to properly study, model, and predict how the Earth will continue to evolve. This is a huge and multifaceted problem, that will undoubtedly require innovations across design, operations, and algorithms. Right now there are so many talented scientists looking to study the Earth and make sure that we can properly steward our planet for future generations, but there are fundamental data bottlenecks that limit what they can do. I want to alleviate these bottlenecks and allow us to better understand the complex world we live in.
What makes you excited about your data science and AI research?
For me, the most exciting aspect of research in fundamental autonomy and robot intelligence is the huge number of doors it opens to better understand the world around us. With every new capability, there is some really compelling environmental or earth science application that now becomes feasible.