Selected Motion Control Problems for a Robotic Guide Dog

Abstract

This paper presents the design and implementation of a vision-based system enabling a robot to autonomously follow a human in dynamic environments. The proposed solution was developed using the quadruped walking robot Unitree Go2, which provides high mobility, terrain adaptability, and dynamic stability suitable for indoor and outdoor operation. The perception subsystem is based on a modified vision setup utilizing the ZED2i stereo camera. The camera enables real-time acquisition of RGB images and depth information, allowing for three-dimensional human detection and localization. Depth data are used to estimate the relative position and distance of the tracked person with respect to the robot, ensuring robust target tracking even in partially cluttered environments. The computational layer is implemented on the embedded platform NVIDIA Jetson Orin Nano, which provides sufficient GPU-accelerated processing power for real-time image analysis and decision-making. The control architecture is built upon ROS 2, enabling modular system integration, efficient inter-process communication, and scalability. Dedicated ROS 2 nodes handle image acquisition, human detection, depth processing, trajectory generation, and motion control. The developed system integrates perception and locomotion control to generate smooth and stable motion commands that allow the robot to maintain a safe and consistent distance from the human target. The overall architecture emphasizes real-time performance, modularity, and robustness, making it suitable for applications such as personal assistance, inspection, and collaborative robotics.