轍認識を利用した小型月面探査ローバーのRGB-D 自己位置推定

抄録

Recently, the popularity of research and development surrounding robotic planetary exploration has drastically increased with many new private entities challenging space exploration. Accompanying this new trend, there has been several major restrictions, one them being a considerable decrease in rover size and weight. With this in mind, the Space robotics lab of Tohoku university has developed a new prototype, named Koguma, which is a 1 kg, three wheel rover. In this work, we describe a novel approach to achieve localization for micro-rovers in environment where the features are scarce. To overcome the issues related with the barrenness of the terrain, we propose a method in which we do not examine what lies ahead and instead look back and apply computer vision techniques to the tracks left by the rover in the soft soil, similar to what it would encounter during a real mission. We have implemented and tested various combinations of feature detectors and descriptors to compare their efficiency and determine the best method for our application.