Embedded Deep Learning Solution for Person Identification and Following with a Robot

This research presents a robust embedded system for following a person, making use of a pipeline of convolutional neural networks. Besides, it features an optical tracking system for supporting the inferences of the neural networks, allowing to determine the position of a person using an RGBD camera. The system is deployed using ROS, and runs in a NVIDIA Jetson TX2, an embedded SoM (System-on-Module), capable of performing computationally demanding tasks onboard, and coping with the complexity required to run a robust tracking and following algorithm. The board is attached to a robotic mobile base, which receives velocity commands to move the system towards the target person.