Leader following: A study on classification and selection

• Leader selection method using prediction of goals of candidates.
• Leader selection using machine learning, trained with real data.
• Study of features contribution in the classification process.
• Experiments with a real robot.

This work proposes a different form of robotic navigation in dynamic environments, where the robot takes advantage of the motion of pedestrians, in order to improve its own navigation capabilities. Instead of treating persons as dynamic obstacles that should be avoided, here they are treated as special agents with an expert knowledge on navigating in dynamic scenarios. This work proposes that the robot selects and follows leaders, in order to move along optimal paths, deviate from undetected obstacles, improve navigation in densely populated areas and increase its acceptance by other humans. To accomplish this proposition, two novel approaches are developed in the area of leader selection. In the first, a motion prediction approach is used, to detect candidates that are moving to the same place that the robot is. In the second, a machine learning algorithm is trained with real examples and is used to select the best leader among several candidates. Experiments with a real robot are performed to validate the proposed approaches.