Power-constrained robots, either those with miniature form factors or long-duration requirements, spend a similar magnitude of energy on computational tasks as they spend actuating a movement, making traditional depth sensors such as LiDARS infeasible. As such, the first objective of my project is to enable perception on power-constrained robots in the LEAN Lab using a deep neural network (DNN) and single RGB camera; this entails producing depth images using the DNN and RGB images, estimating uncertainties of the DNN’s predictions, and on-the-fly training of the DNN. The second phase will focus on adaptively adjusting the FPS at which an onboard sensing network runs, depending on the level of certainty regarding the environment, to conserve energy.

Through this SuperUROP, I want to develop my research, communication, and writing skills, as well as experience what a more structured research process is like. Having done coursework as well as extracurriculars relating to robotics and controls, I’m excited to explore the field further in-depth, especially with regards to path-planning and energy efficiency, and contribute meaningfully to my group.