Topological map for biologically inspired vision based navigation

Kurzfassung

As a step towards achieving a mobile robot which can navigate autonomously, this work aims to develop a method to create a topological map for biologically inspired viewframe based navigation. The developments made in computer vision and the reducing cost of cameras have lead to research efforts in appearance based mapping and navigation for mobile robots. The idea to use biological navigation concepts is inspired from the reason that insects like ants and bees are able to navigate autonomously and robustly in their environments mostly using vision as their primary sensor in spite of having limited computational abilities and a small brain. This can come in handy for small robots like flying robots which are limited in computational and memory resources. Locations in the environment have been represented using a viewframe that is identified by visually inspecting the environment at a particular location and consists of a set of landmark views where each landmark view denotes a landmark observation (ID, descriptor and angle). BRISK features extracted from omnidirectional images were used to detect and identify these landmarks in the environment. A method to build a topology map by using dimensionality reduction techniques was implemented and evaluated using this viewframes information. Two measures namely overlap similarity measure and bearing angle dissimilarity measure have been implemented and used to represent the dissimilarity between any two viewframes. Using these measures, a dissimilarity matrix was computed and provided as input to the dimensionality reduction techniques. Also, two quality measures namely the connectivity measure and isometry measure were introduced for evaluating the generated map quality. Experiments were conducted in simulation for a set of robotic trails under varying sensor and environmental parameters. Also, experiments were conducted on the Pioneer 3-DX robot in an indoor lab environment. Results show the success of the approach in building a topological map automatically using only landmarks information. The resultant map was used for homing experiments and was successful in identifying shorter and optimum paths as compared to its previous version that is Trail-Map which only preserves relationship between adjacent locations.