Advancing Vision-Guided Autonomous Ground Vehicles: Deep Learning Applications for Improved Perception and Specialized Industrial Operations

The performance of vision-guided ground robot vehicles is in fact a critical aspect in various fields of application. These robots frequently rely on visual sensors to perform the given task, to interact with their environment and to navigate to specified target locations. Deep learning techniques have been increasingly adopted to improve the monitoring performance of autonomous ground robots. However, the deployment of these models on autonomous ground robots remains an unexplored area of research.
This thesis introduces novel applications for vision-guided robots across various industries. It aims to clarify the benefits of applying deep learning techniques to augment the performance of robotic operations tailored to specific applications. The proposed techniques make possible the use of ground robot vehicles in more specialized applications, such as road maintenance inspection, construction inspection, and garbage segregation.
The study presents a step-by-step implementation of AI applications in the field of robotics. By starting from scratch with dataset creation, which includes gathering and annotation of data, it enables data-driven decision-making for specific applications. This approach ensures that the vision-guided robotic systems are trained on relevant and representative data, leading to more accurate and effective performance in real-world scenarios. In addition, by integrating semantic information into the 3D representation, the approach significantly enhances the performance of autonomous navigation. This integration provides a more comprehensive understanding of the environment, enabling the autonomous system to make more informed and accurate decisions.
This research contributes to the advancement of robotics and AI applications, paving the
way for more efficient and specialized robotic operations.