Robot vision network based on ceiling map sharing

Abstract

In the system introduced in this report, a ceiling map is shared by mobile robots inside a building. Firstly, the map is generated from a video trough an "upward" camera on a leading robot. Next, location of a following robot is estimated and plotted on the map. If there is debris on the floor, the estimated location contains error. To avoid this "estimation" problem, in our previous report, we extracted motion information from a "forward" camera and fed it back to the location estimation. However, it requires transmitting video signals from two cameras - upward and forward. It means that data amount for transmission via digital network is doubled. To avoid this "transmission" problem, in this report, 1) we remove redundancy on motion information between the two cameras. For further improvement, 2) we assign much load to the leading robot than following robots. As a result, data amount for communication among mobile robots is significantly reduced. It contributes to have more following robots under limited data transmission capacity of a digital network.