Určení bezpečnostního prostoru pro kooperaci člověka s robotem pomocí počítačového vidění a AI

Abstract

The dissertation thesis focuses on the design and implementation of a system for determining safety and comfort zones during human interaction with a mobile robot, using computer vision and artificial intelligence methods. The aim of the work is to enhance the safety and naturalness of human–robot interaction in dynamic environments, where traditional approaches lack sufficient adaptability. The thesis introduces a custom proxemic model based on the golden ratio, which divides a person’s personal space into finely graded zones and considers the person’s relative position, speed, and direction of movement in relation to the robot. The solution includes person detection, tracking, and classification by age and gender using neural networks, human pose estimation, and the calculation of individualized zones for each detected person. For classification purposes, a custom dataset was created, combining real and synthetically generated images. The entire system was implemented as an application on an NVIDIA Jetson Xavier NX computing unit, using a ZED 2i depth camera, and was tested on two types of mobile robotic platforms (Unitree Go1 and Agilex Scout Mini). The system’s functionality was validated through an experiment that utilized subjective interaction assessments via a questionnaire inspired by the Godspeed Questionnaire Series. The results showed that the proposed system is perceived as safe and comfortable, confirming the suitability of the applied methodology for defining zones in physical interaction contexts. The work offers contributions both to the scientific research of proxemics in human-robot interaction and to its practical applicability in the field of service robotics.