Computer vision algorithms for autonomous mobile robot map building and path planning

This paper describes a new algorithm for autonomous mobile robot (AMR) map building and path planning in an unknown environment. The robot learns its environment using a view-based probabilistic interpretation of the visible corner and edge features. No attempt is made to locate physical world features in a traditional Euclidean coordinate system, rather the algorithm produces map building through measures of visual similarity alone. A robust map building algorithm is developed, whose robustness is notably, independent of the size of the learnt environment. This enables large environments to be learnt reliably. To date no other AMR map building algorithm is capable of robustly learning an arbitrary sized unknown environment. The robot performs robust intra-location localization and interlocation navigation using an internal representation of the spatial arrangements of nearby known locations