Soumya Sudhakar
Soumya Sudhakar is a PhD candidate in aeronautics and astronautics whose research is focused on decision-making algorithms for miniature or long-duration robots that are energy- efficient in actuation and computing. As a MathWorks Fellow, she will investigate new methods for measuring uncertainty in machine learning in a computationally efficient way. To address the high computational costs of current methods, Soumya has proposed a novel method to exploit temporal correlation across the inputs such that only a single inference is required per input, for much greater computational efficiency. She is also investigating efficient online learning, an essential problem for robots with limited resources in the era of extremely large AI models, as well as the topic of computing’s climate impact, an important subject given the large amount of computing required for the navigation of autonomous vehicles. Soumya’s research makes significant use of MATLAB, and her research offers exciting new directions in energy-constrained perception, planning, and actuation to advance miniature and long-duration robotics in novel applications.