A robot can accurately localize itself and navigate in an indoor environment based on information about the operating environment, often called a world or a map. While typical maps describe structural layouts of buildings, the accuracy of localization is significantly affected by non-structural building elements and common items, such as doors, appliances, and furniture. This study enhances the robustness and accuracy of indoor robot localization by dynamically updating the semantic building map with non-structural elements detected by sensors. We propose modified Adaptive Monte Carlo Localization (AMCL), integrating object recognition and map updating into the traditional probabilistic localization. With the proposed approach, a robot can automatically correct errors caused by non-structural elements by updating a semantic building map reflecting the current state of the environment. Evaluations in kidnapped robot and traditional localization scenarios indicate that more accurate and robust pose estimation can be achieved with the map updating capability.

한국항공운항학회지(Journal of the Korean Society for Aviation and Aeronautics)

Deep reinforcement learning (DRL) for autonomous mobile robot navigation faces several inherent limitations. The stochastic nature of actions generated by DRL policies can undermine performance consistency, while inefficient exploration frequently delays the learning process or prevents the discovery of optimal solutions. This research aims to enhance the robustness of path planning by addressing these challenges. To achieve this goal, we propose a hybrid approach that integrates the flexible decision-making capabilities of deep reinforcement learning with the stability of traditional path planning. The proposed model adopts the Twin Delayed Deep Deterministic Policy Gradient (TD3) network as its base. Notably, we pre-process LiDAR point cloud data to extract only essential features for the state representation, thereby preventing performance degradation from high-dimensional inputs and improving computational efficiency. Our model optimizes the learning process through two core strategies. First, it prioritizes experience data generated during training based on negative rewards, guiding the model to learn more frequently from critical failures rather than redundant successes. Second, it dynamically compares the action proposed by the TD3 network with a goal-oriented action from a classical path-planning algorithm in real time. By selecting the action with the higher estimated value, the model guides the policy toward a stable and effective trajectory from the earliest stages of training. To validate the efficacy of our approach, we conducted simulation-based experiments comparing the performance of the proposed model with existing reinforcement learning networks. To ensure statistical significance and mitigate the impact of random initialization, all reported results are averaged over 10 independent runs with different random seeds. The results quantitatively demonstrate that our model achieves significantly higher and more stable reward values, confirming a robust improvement in the path-planning process.

한국항공운항학회지(Journal of the Korean Society for Aviation and Aeronautics)

This paper presents a field-deployable autonomous 3D scanning framework capable of self-decision-making in unknown, cluttered outdoor environments. The proposed system introduces a two-leve... Practical ApplicationsThis paper documents a field-deployable workflow for autonomous 3D scanning in cluttered, unknown outdoor sites such as disaster sites and active construction areas. The syste...