RBE Masters Thesis Presentation
Karter Krueger
Multi-Level Semantic SLAM
Tuesday, April 25th
10:30 AM - 12:00 PM
Zoom Link: https://wpi.zoom.us/j/91471216151
Abstract: This thesis presents a novel Multi-Level Semantic Simultaneous Localization and Mapping (ML Semantic SLAM) method designed to improve the accuracy and reliability of robot mapping and localization in sparse and repetitive environments. By incorporating both low-level and high-level semantic features, our approach addresses the challenges associated with limited distinguishable keypoint ORB features and generates a more semantically rich map. We detail the development, implementation, and evaluation of our ML Semantic SLAM system in various simulated and real-world environments, demonstrating its superior performance compared to traditional SLAM techniques, such as ORB-SLAM2. The system achieves up to a 70% error reduction in highly sparse environments and exhibits modest improvements in more moderate environments, showcasing its robustness and versatility. We also propose several future directions to extend the research. By continually refining and expanding upon the Multi-Level Semantic SLAM method, we hope to enable more accurate and reliable SLAM systems for various real-world applications.
Advisor:
Professor Jing Xiao, Worcester Polytechnic Institute (WPI)
Committee:
Professor Carlo Pinciroli, Worcester Polytechnic Institute (WPI)
Professor Nitin Sanket, Worcester Polytechnic Institute (WPI)