Adrian-Paul Botezatu

This work proposes a hybrid deep learning-based framework for visual feedback control in an eye-in-hand robotic system. The framework uses an early fusion approach in which real and synthetic images define the training data. The first layer of a ResNet-18 backbone is augmented to fuse interest-point maps with RGB channels, enabling the network to capture scene geometry better. A manipulator robot with an eye-in-hand configuration provides a reference image, while subsequent poses and images are generated synthetically, removing the need for extensive real data collection. The experimental results reveal that this enriched input representation significantly improves convergence accuracy and velocity smoothness compared to a baseline that processes real images alone. Specifically, including feature point maps allows the network to discriminate crucial elements in the scene, resulting in more precise velocity commands and stable end-effector trajectories. Thus, integrating additional, synthetically generated map data into convolutional architectures can enhance the robustness and performance of the visual servoing system, particularly when real-world data gathering is challenging. Unlike existing visual servoing methods, our early fusion strategy integrates feature maps directly into the network’s initial convolutional layer, allowing the model to learn critical geometric details from the very first stage of training. This approach yields superior velocity predictions and smoother servoing compared to conventional frameworks.

This work proposes a hybrid deep learning-based framework to visual feedback control an eye-in-hand robotic system. The framework uses an early fusion approach in which real and synthetic images define the training data. The first layer of a ResNet-18 backbone is augmented to fuse interest-point maps with RGB channels, enabling the network to capture scene geometry better. A manipulator robot with an eye-in-hand configuration provides a reference image, while subsequent poses and images are generated synthetically, removing the need for extensive real data collection. The experimental results reveal that this enriched input representation significantly improves convergence accuracy and velocity smoothness compared to a baseline that processes real images alone. Specifically, including feature point maps allows the network to discriminate crucial elements in the scene, resulting in more precise velocity commands and stable end-effector trajectories. Thus, integrating additional, synthetically generated map data into convolutional architectures can enhance the robustness and performance of the visual servoing system, particularly when real-world data gathering is challenging.

The rapid advancement of autonomous vehicle technology has brought into focus the critical need for enhanced road safety systems, particularly in the areas of road damage detection and surface classification. This paper explores these two essential components, highlighting their importance in autonomous driving. In the domain of road damage detection, this study explores a range of deep learning methods, particularly focusing on one-stage and two-stage detectors. These methodologies, including notable ones like YOLO and SSD for one-stage detection and Faster R-CNN for two-stage detection, are critically analyzed for their efficacy in identifying various road damages under diverse conditions. The review provides insights into their comparative advantages, balancing between real-time processing and accuracy in damage localization. For road surface classification, the paper investigates the classification techniques based on both environmental conditions and material road composition. It highlights the role of different convolutional neural network architectures and innovations at the neural level in enhancing classification accuracy under varying road and weather conditions. The main finding of this work is that it offers a comprehensive overview of the current state of the art, showcasing significant strides in utilizing deep learning for road analysis in autonomous vehicle systems. The study concludes by underscoring the importance of continued research in these areas to further refine and improve the safety and efficiency of autonomous driving.

A visual servoing system is a type of control system used in robotics that employs visual feedback to guide the movement of a robot or a camera to achieve a desired task. This problem is addressed using deep models that receive a visual representation of the current and desired scene, to compute the control input. The focus is on early fusion, which consists of using additional information integrated into the neural input array. In this context, we discuss how ready-to-use information can be directly obtained from the current and desired scenes, to facilitate the learning process. Inspired by some of the most effective traditional visual servoing techniques, we introduce early fusion based on image moments and provide an extensive analysis of approaches based on image moments, region-based segmentation, and feature points. These techniques are applied stand-alone or in combination, to allow obtaining maps with different levels of detail. The role of the extra maps is experimentally investigated for scenes with different layouts. The results show that early fusion facilitates a more accurate approximation of the linear and angular camera velocities, in order to control the movement of a 6-degree-of-freedom robot from a current configuration to a desired one. The best results were obtained for the extra maps providing details of low and medium levels.

This survey explores the evolution and applications of Visual Servoing Systems in robotics, emphasizing the transition from traditional image processing techniques to the incorporation of neural networks for feature extraction and control. Robotic systems, integral to manufacturing, surveillance, and healthcare, increasingly rely on Visual Servoing for enhanced interaction within various work environments. Initially focused on auxiliary sensors like visual sensors for robustness and accuracy, recent advances have seen a shift towards integrating deep learning methods for direct control and feature extraction. The survey covers the differences that emerge from classical Visual Servoing architectures to novel methods involving Deep Learning, highlighting their respective advantages and limitations regarding stability, precision, and real-time applicability. Innovative approaches, such as Direct Visual Servoing and the use of siamese networks for camera position estimation, demonstrate significant progress in overcoming the challenges of traditional Visual Servoing. Through detailed examination of leading research, the survey highlights the potential of neural networks to revolutionize this domain by enhancing feature extraction, reducing reliance on precise calibration, and improving control laws for complex robotic tasks.