Visual Navigation in Unstructured Environments and the Development of a Multi-purpose Mobile Robot Platform

In this paper, we present the development of a visual navigation and mobile robot platform designed for autonomous driving in outdoor unstructured environments. To address the challenges posed by outdoor environments, where inter-object features are ambiguous and environmental conditions are highly irregular, we apply a semantic segmentation technique to the mobile robot’s local navigation system, enabling it to move within a navigable region. A simplified segmentation structure is incorporated for processing large volumes of visual data and is integrated with vision-based control strategies to facilitate effective navigation planning. Additionally, we propose a supplementary method that involves road segmentation, which uses depth information to ensure stable and robust driving. Our research also includes the design of a wheeled mobile robot capable of operating in various environments, highlighting its practical applicability across diverse fields. The potential of this platform is validated through empirical evaluations conducted with self-developed real robots in various driving scenarios, demonstrating over 80% accuracy in navigable region classification and over 90% accuracy in road segmentation performance under dynamic conditions.