Localization is very helpful for goal-oriented navigation of a mobile robot. In this article, we describe the challenges we faced when designing a low-cost indoor localization system that can be employed on consumer and domestic robots for the systematic navigation in household environments. Our system uses active beacons that project a pair of infrared (IR) spots onto the ceiling and a sensor on the robot which observes them. In order to reduce cost, we designed a 3-diode sensor, i.e. a three pixel camera, that measures directions to the spots. In contrast to e.g. a typical camera with VGA or higher resolution that would allow for a precise tracking of the beacon spots, our low-cost sensor suffers from multi-path, i.e. light not only reaches the sensor directly but also through reflections from walls and other furniture. We tackle this problem by a localization method that learns the light distribution in the room through a simultaneous localization and mapping (SLAM) approach. Our experiments provide numbers on the accuracy and consistency of this method. The presented system is implemented on our Mint robot equipped with an ARM 7 processor and 64 kByte of RAM for the autonomous cleaning of floors.
展开▼
