Optimizing Data Capture Through Object Recognition for Efficient Sensor and Camera Management with a Quadruped Robot

Abstract

This paper explores the integration of object recognition to advance autonomous mobile robotics, with a specific focus on employing sensor fusion in quadrupedal swarm units, exemplified by the Unitree GO1 robotic dog. The proposed system utilizes an RGB camera and the YOLOv8 machine learning module to enable the robotic dog to identify objects, subsequently triggering predefined actions or activating additional hyper-spectral sensors and cameras. This approach enhances the robot's environmental interaction, allowing it to optimize power efficiency by activating sensors selectively, resulting in a more focused dataset. The technology's applications span various domains, including threat detection in home defense, surveillance of civic structures, and monitoring industrial components. The paper delves into the customization potential through the creation of bespoke datasets and model training, incorporating techniques such as transfer learning and domain adaptation to tailor the system to user-specific requirements. Beyond its immediate applications, the implications of object recognition and AI-assisted robotics extend to diverse scientific communities, offering a versatile tool for a wide array of applications.