Experiments in leader classification and following with an autonomous wheelchair

Abstract

With decreasing costs in robotic platforms, mobile robots that provide assistance to humans are becoming a reality. A key requirement for these types of robots is the ability to efficiently and safely navigate in populated environments. This work proposes to address this issue by studying how robots can select and follow human leaders, to take advantage of their motion in complex situations. To accomplish this, a machine learning framework is proposed, comprising data acquisition with a real robot, data labeling, feature extraction and the training of a leader classifier. Preliminary experiments combined the classification system with a multi-mode navigation algorithm, to validate this approach using an autonomous wheelchair.