Vision-Based Object Sorting in Dynamic Environments using YOLO for RoboCup ARM Challenge 2023

In the intersection of Artificial Intelligence (AI) and Robotics, autonomous object manipulation has emerged as a pivotal research domain, finding applications in sectors as diverse as manufacturing and healthcare. While significant strides have been made in object identification, trajectory planning, and manipulation tasks, these advancements often fall short in real-world, dynamic settings, revealing a critical research gap in algorithmic adaptability and computational efficiency. This paper addresses the gap by applying YOLOv4, a state-of-the-art object detection AI algorithm, within the context of the RoboCup Autonomous Robot Manipulation (ARM) Challenge—a competitive platform designed to simulate real-world complexities in autonomous systems. The research objectives are twofold: firstly, the performance of YOLOv4 in object detection and classification under dynamically changing conditions is rigorously evaluated, employing varied training strategies in both simulated and real-world scenarios. Secondly, the robustness and adaptability of the algorithm in object manipulation tasks are scrutinized and facilitated by the real-world dynamics inherent to the competition. Empirical results demonstrate the algorithm’s high accuracy, efficiency, and robustness in dynamic environments, substantiating its broader applicability in the development of autonomous systems.