The research objective of this project is to further the state-of-the-art in simultaneous localization and mapping (SLAM) methods for robotic vehicles. SLAM is the process of building a map of the surrounding vehicle environment while simultaneously localizing the vehicle relative to the map under construction. This work is motivated by cooperative control of small, agile, aerial robotic systems that are envisioned to execute tasks such as the transportation and manipulation of payloads using grippers or cables, assembly and construction of structures, and continuous surveillance and reconnaissance. The sensors these vehicles typically use are corrupted by bias and high-frequency noise. On-board computers are computationally limited as well. A computationally simple yet effective online SLAM algorithm that is immune to bias and noise typical of on-board sensors is crucial to the proliferation of autonomous robotic systems, and the focus of the proposed research.
This project focuses on theoretical advancements to DG-SLAM that will make it more practicable. Specifically, (a) how to account for measurement bias, and (b) how to use vector measurements directly, will be the focus of the proposed research. These two extensions of DG-SLAM are critical to the robustness, reliability, and ease of implementation of DG-SLAM on real-world vehicles.
