RF-guided exploration for robotic manipulation

Abstract

This thesis presents the design, implementation, and evaluation of RVExp, a system that leverages radio perception to design efficient exploration controllers in highly occluded and non-line-of-sight settings. RVExp relies on eye-in-hand vision sensor and battery-less RFID tags attached to the object of interest. Since RF (radio frequency) signals can traverse everyday occlusions, RVExp uses the RFIDs' signals to locate an object through occlusions. At the heart of RVExp's design is a novel radio-visual planner that uses the RFID's location to semantically explore the environment and plan an efficient trajectory toward an occluded target object. The controller can efficiently discover obstacles by biasing a Gaussian belief space planner using the RFID's location. I have built RVExp end-to-end and evaluated its performance in challenging real-world environments. The results show the feasibility of RF-guided exploration and its superiority over standard exploration.