Trajectory learning for a friendly interactive robot for service tasks

To perform missions, indoor mobile robots must be able to follow collision-free trajectories. We have designed a learning system to teach the robot trajectories so that the feasibility of the paths can be ensured. During the execution, the robot follows the learned trajectory but online localization and obstacle avoidance are also performed so that the robot can always achieve the requested mission even in indoor dynamic environments. The learning process is made of three phases. The first one is a structuring phase in which we classify all trajectories to be learned. The next phases are the teleoperation phase for data acquisition and the checking phase to test the learned trajectory. We present all the phases and some experiments