Dense topological maps and partial pose estimation for visual control of an autonomous cleaning robot

In this article the author presents a navigation strategy for household cleaning robots which leads the robot along parallel and meandering lanes in order the cover rectangular surfaces. This strategy is a topological map of the environment which are graph-based representations of the environment. The preference goes to this strategy due to 1. building these topological maps from omnidirectional images is possible in real-time even with limited computational resources (domestic robots often are primitive) 2. The trajectory controller can use existing local visual homing algorithms for estimating spatial relations between place nodesn 3. additional task-relevant information can be attached to the place nodes.

Additionally different types of topological maps are presented, from standard to dense topological maps which are finer in resolution. One of the setbacks of the strategy is that only the robots distance to the previous lane and current orientation are computed. So an estimation of robot’s position is not made. Due to this, the robot requires necessary and sufficient information, and if this is not available, the robot will not be able to accomplish its task.