Vision-based UAV-AGV collaborative system is one of the popular collaborative robotic systems which is advantageous in GPS (Global Positioning System) inaccessible areas for various applications. Inspection, surveillance, target tracking and environment mapping are the potential applications of the system. UAV can assist the navigation of AGV and AGV can support the long range missions of UAV. UAV carries a down facing camera for the relative localization of AGV. This configuration is the most popular vision-based UAV-AGV system. Collaborative motion between both agents is an important aspect of the system to perform any task. An image plane based autonomous waypoint navigation technique for the collaborative motion of UAV is developed in this paper. The collaborative task of UAV is to visit the various target positions in the image plane (while keeping AGV in field of view (FOV)) to gain the knowledge of the surroundings of AGV or to perform inspection at the target positions. Each target position is a waypoint to be reached by the UAV in the image plane. UAV plans minimum acceleration trajectories connecting each target location and reaches one by one. A sliding mode theory based novel kinematic controller is designed for tracking of the designed trajectories. The proposed method is validated in the Gazebo simulation tool. The method is useful in GPS denied environments for the UAV to plan smooth navigation to visit multiple positions in the image plane. UAV can reach various positions in sequence in the image plane to increase the knowledge of the environment. The method is also useful to perform discrete inspection of the horizontal infrastructures which are not reachable to AGV during collaborative inspection task.
The Deep Neural Network (DNN), especially Convolutional Neural Networks (CNN), has revolutionized image pattern recognition by utilizing filters for feature extraction. Achieving superior classification results across diverse datasets relies heavily on identifying the optimal DNN structure and associated hyperparameter selection. However, determining this optimal configuration often requires significant computational resources and time. This paper presents a Multi-Fidelity Design Optimization (MFDO) approach, which quantifies deep learning structures and hyperparameters using the Genetic Algorithm (GA) optimization method. The MFDO efficiently explores the design space across various DNN hierarchical structures to attain optimal solutions. To reduce optimization time, the paper employs GA optimization on a smaller image dataset, followed by iterations involving training and cross-validation on a larger dataset. The DNN structure includes multiple modules with diverse choices such as convolutional layers, pooling layers, dropout layers, or fully connected layers, all optimized through encoded GA chromosomes. Training and validation are conducted using the MNIST and Fashion MNIST datasets. Results demonstrate the effectiveness of the GA process with the "low-fidelity dataset" in achieving shorter computational times, while validation with the "high-fidelity dataset" confirms the accuracy of the proposed design. The MFDO process effectively determines the optimal DNN structure with the best hyperparameters.
Research for the optimization problem when designing mechanical system is often time consuming and may require numerous efforts in finding the optimal parameters. In this study, an example of finding optimal parameters of the Passive Dynamic Vibration Absorbers (PDVA) of the ropeway carriage system is proposed. By using the time domain response for auto-tuning, this experiment aims at providing an fast and intuitive method for engineers to fast prototyping using simulation results. The experiment demonstrate that this concept is successful, and suitable for future applications for systems with similar dynamic response.
Marine robotics plays a fundamental role in executing various underwater complex missions. In this scenario tethered Remotely Operated Vehicles (ROVs) offer benefits such as efficient data transmission and strong physical connection for emergency cases, but their motion is confined by the length of the tether. This study addresses the issue of extending the range of the ROV by implementing a new control strategy cooperating with an Autonoumous Surface Vehicle (ASV). The proposed system ASV-ROV confronts two main challenges: the constrained movement of the ROV and the risk of cable entanglement. Therefore, a new control system is built to provide the ROV a smooth motion and avoid situations that are likely to generate cable entanglements. The new control system enables the ASV to align and to keep a certain distance from the ROV, regulating at the same time the released cable length in the water. The paper addresses the design of ASV-ROV cooperative system, the modelling of each vehicle, the control strategy, and simulation results. Future work research is also outlined at the end.
The objective of this study is to identify the optimal object detection architecture for training on a specific type of defect detection, namely incorrectly polished surfaces on aluminium elements. In order to facilitate a meaningful comparison of the various architectures, a maximum training time of approximately one hour was established for each architecture. Using the Darknet framework and a specific dataset, five architectures were compared (for the time being). The parameters of the various architectures, including network size, number of batches, and so forth, were modified according to a well-defined and systematic procedure. The preliminary findings indicate that the YOLOv4-tiny network exhibits superior training performance on this dataset, rendering it an optimal choice for industrial applications. This research provides support to small and medium-sized enterprises (SMEs) by identifying effective object detection architectures for quality control and highlighting avenues for advancing AI-driven defect detection in manufacturing.
