The autonomous operation of mobile robots benefit from a detailed metric and semantic understanding of their environment. Typical mapping approaches rely heavily on depth measured directly by sensors or estimated with parallax. These depth values typically become unreliable at far distances, so mapping is often limited to the robot’s local area. We utilize frame-to-frame tracking and cross-observation association to sparsely localize far-field landmarks. Each landmark is attributed an open-set semantic descriptor and location range allowing them to be used for high-level planning.

My SuperUROP project will give me the chance to contribute towards more capable and robust robotic systems. I look forward to developing and testing my work, and pushing towards more useful robots for the future.