Robust Map Learning with Low-Cost Sensors for an Autonomous Robot

The paper describes several approaches and experimental results for learning a map of an indoor environment, using a combination of odometry, sonar range sensors, and vision. The aim of the presented research work is consistent real-time map-learning in indoor-environments with low-cost sensors and limited computational resources and without installations in the environment itself. Therefore, we cannot afford high-complexity techniques like probabilistic SLAM. Instead, our approach involves a set of algorithms for odometry correction and automatic recalibration of position estimation. Results from different environments are presented that show that the methods we developed are appropriate for mapping large scale areas by means of low-cost sensors.