Development of an indoor topological navigation framework for mobile robots using 360-degree cameras

Date

2025-4-7

Author

Oydu, Onur

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

110
views

0
downloads

Cite This

This thesis presents the development of an innovative topological navigation framework for mobile robots operating in indoor environments. A metric-free vision-based navigation system using 360-degree cameras is proposed to overcome the limitations of metric-based navigation systems. Equirectangular images are taken at each node by using a 360-degree camera mounted on the robot head. A topological map with nodes defined by equirectangular images is generated, and connections between nodes are determined using matching features. Feature extraction (ALIKEd) and matching (LightGlue) algorithms identify matching features. The 8-point algorithm with RANSAC estimates the heading vector and relative orientation between nodes. Heading vectors and relative orientations produce scale-free node coordinates, which are then validated with the reference floor plan. A graph search algorithm is utilized in the path-finding stage. Four routes across the map were driven in the normal direction and then in reverse, confirming that the proposed topological navigation algorithm supports reliable bidirectional travel. Also, another topological map consisting of fully connected 9 nodes with known positions was generated. Every node was used as a start once and as a goal for each of its eight neighbors, giving 72 start–goal trials with a mean position error of 0.32 m and an overall success rate of 88.9 %. A hand-mounted 360-degree camera performs vision-only triangulation of a specific object in the goal node. Two routes were traversed forward and reverse, confirming reliable bidirectional approach to the object. Also, the triangulation algorithm was validated in three indoor layouts, achieving a mean position error of 4.37 cm and a mean orientation error below 2.5 degrees. Overall, the results demonstrate that once the robot reaches its target pose, the manipulator can servo precisely to the object while remaining entirely within its 0.87 m horizontal workspace.

Subject Keywords

Topological navigation, Equirectagular image, Feature-matching, 8-point algorithm, Metric-free navigation

URI

https://hdl.handle.net/11511/114587

Collections

Graduate School of Natural and Applied Sciences, Thesis