Indoor Mapping with Machine Learning Algorithm using Khepera III Mobile Robot

Abstract

Small robots can be beneficial in many applications as they have the advantage of reaching small spaces. For these robots to be truly autonomous, ability to map their surrounding is essential. Accuracy of mapping is related closely to sensor’s precision. However, small robots can only be equipped with small sensor that is typically has noisy characteristic with cheaper cost, such as sonar sensor and infrared sensor. To enhance the quality of map build by noisy and low-cost sensor, machine learning algorithm integration is a good approach. In this work, multiple learners, which are Naïve Bayes, Decision Tree, Neural Network and AdaBoost, were experimented with occupancy grid map algorithm using Khepera III robot platform. Then, the results of their fitness score according to the maps build were compared. The results show that Neural Network performed the best with the occupancy grid map algorithm.