Biologically Inspired Autonomous Robotic Navigation Using High-Level Object Detection
Abstract
Caleb Hulbert B.S., Cornell University, 2016 Biologically Inspired Autonomous Robotic Navigation using High-level Object Detection Thesis advisors: Damian Lyons, Ph.D., Daniel Leeds, Ph. D. Autonomous navigation behaviors are observed in almost every animal, including humans. Neurological and behavioral studies alike seek to deconstruct biological navigation into its most basic components. In this thesis, we propose a method for autonomous navigation that makes use of high-level visual features. Namely, we focus on navigation limited to using visual input and associated semantic information. This thesis seeks to show that human-inspired object recognition, when applied in robotics, can provide an effective method for autonomous navigation. We make use of Gazebo robotic simulation software, the ROS robotic operating system, Turtlebot robot software, YOLO (You Only Look Once) object recognition, and SIFT feature extraction to compare navigation results between two methods: The proposed autonomous navigation method, referred to here as “semantic navigation” was compared to a contemporary navigational method known as average-landmark-vector (ALV) navigation using SIFT features as landmarks. The experiment consisted of simulated traversals of randomly chosen paths from a Gazebo world, each traversal consists of multiple place-to-place transitions, whose data have been recorded and analyzed. Results show that semantic navigation outperforms ALV when comparing the accuracy of the end location with respect to the goal, the number of movement iterations required to complete a transition, and the amount of extraneous distance traveled during each transition. These metrics support Semantic Navigation as a viable alternative which outperforms ALV navigation in certain goal-oriented visual homing tasks.
Subject Area
Robotics
Recommended Citation
Hulbert, Caleb, "Biologically Inspired Autonomous Robotic Navigation Using High-Level Object Detection" (2019). ETD Collection for Fordham University. AAI10745648.
https://research.library.fordham.edu/dissertations/AAI10745648
