Journal of Field Robotics最新文献

Although season-long cotton flowering time characterization has value to breeders and growers, a manual data collection process is too laborious to be practical in most cases. In recent years, several fully automated flower counting approaches have been proposed. However, such approaches are typically designed to run offline and require a significant amount of computation. Furthermore, little thought has gone into developing convenient interfaces and integrations so that a layperson can use such systems without extensive training. The goal of this study is to develop a flower tracking system that is deployable on a ground robot and can operate in real time. A previous GCNNMatch++ approach was modified to increase the inference speed. Additionally, data from multiple cameras were fused to avoid canopy occlusions, and three-dimensional flower locations were extracted by integrating GPS data from the robot. It is shown that the approach significantly outperforms UAV-based counting and single-camera counting while running at above 40 FPS on an edge device, achieving a counting error of 15. Overall, it is believed that the highly integrated, automated, and simplified flower counting solution makes significant strides toward a practical commercial cotton phenotyping platform.

Autonomous service robots are becoming increasingly common in human-centric, long-term deployments in unstructured indoor environments. Robotic vision is a crucial capability, enabling robots to perceive and interpret high-level environmental features from visual input. While data-driven approaches based on deep learning have advanced the capabilities of vision systems, applying these techniques in real robotic scenarios still presents unique methodological challenges. Conventional datasets often do not represent the object categories that a service robot needs to detect. More importantly, state-of-the-art models struggle to address the demanding perception constraints faced by service robots, posing the need for adaptations to the specific environments in which the robots operate. We devise a method that addresses these challenges by leveraging photorealistic simulations to create synthetic visual datasets from a robot's perspective. This approach balances data quality with acquisition costs, enabling the training of deep, general-purpose detectors tailored for service robots. We then demonstrate the benefits of qualifying a general detector for the domain in which the robot is deployed, studying the trade-off between data-acquisition efforts and performance improvement. We evaluate our method using a representative selection of prominent deep-learning object detectors for the challenge of recognizing, in real time, the presence and traversability of doorways. This task, which we refer to as door detection, is fundamental to numerous significant robotic tasks, such as tracking the changing topology of dynamic environments. We conduct an extensive experimental campaign in the field, considering different real-world setups while emulating the typical challenges encountered in long-term deployments of service robots. Our key findings demonstrate that simulation and qualification techniques can significantly reduce costs associated with domain adaptation for service robots. While simulation allows embedding the robot's perspective during the training of end-to-end robotic vision modules, qualification is essential to improve their robustness over challenging detection instances, thus reaching the performance level typically required by realistic long-term deployments of service robots.

In complex and dynamic environments, the decision-making sequence of individual robots significantly influences the effectiveness of collaboration and cooperation among multi-robot systems in completing tasks. This paper focuses on the division of labor in autonomous multi-robot systems, aiming to find optimal strategies for resource allocation among robots operating in complex scenarios. Each robot makes independent yet interacting decisions in relatively isolated dynamic environments. We propose a model that applies game theory from economics, classifying the robots into resource-providing robots and resource-consuming robots. Resource providers acquire resources and compete to determine the optimal strategy, while resource consumers purchase resources, making decisions based on the pricing set by providers. The problem is formulated as a two-stage game. In the first stage, resource providers engage in resource games, abstracted into Cournot or Stackelberg models, where optimal decisions are made based on available resources and estimated strategies of other participants. The second stage involves price games between providers and consumers, analogous to market supply and demand relationships. Price adjustments and demand changes lead to the discovery of Nash equilibria in the price game. Simulations were conducted to compare the system-wide benefits when providers adopt different strategies in the first stage. Results indicate that using the Stackelberg model yields higher overall benefits, further demonstrating the practicality and effectiveness of the proposed strategies. This highlights the importance of strategic model selection in optimizing the performance and resource efficiency of multi-robot systems operating in dynamic environments.

This paper presents an improved Dung Beetle Optimization (DBO) algorithm assisted by a neural network to solve the inverse kinematics of a 5-DoF (degree-of-freedom) warehouse loading and unloading robot. The neural network was trained on pose data of the robot arm's end-effector and generated predicted joint angles as output. A genetic algorithm was employed to enhance the design of the network. The selected network structure has been demonstrated to reduce the mean Euclidean Distance (ED) of the end-effector positions to below 9 mm. To achieve more accurate inverse kinematics predictions, experimental tuning was conducted to determine the optimal configuration for the DBO algorithm. Then we developed a hybrid model that integrated the neural network and the DBO algorithm. Experimental results indicated that the positioning errors of the DBO and hybrid algorithms were significantly reduced compared to those of the neural network. Moreover, the proposed model reduced the total inference time by 20.7% and the mean ED by 16.4% compared to the DBO algorithm. In trajectory planning validation experiments, all fitting errors were less than 5 mm, thereby meeting the practical requirements for warehouse handling. Therefore, the proposed neural network-assisted evolutionary algorithm outperforms both the neural network and DBO algorithms, providing a fast and accurate solution to the inverse kinematics problem of the warehouse manipulator.

The multi-filament carbon fiber winding technique is an advanced and novel process, where tension control and regulation during winding play a crucial role in determining product performance. Based on this, this paper introduces fractional-order modeling to derive the physical model of the fiber winding process and optimizes the design of the traditional fractional-order proportional-integral-derivative (FOPID) controller, to obtain the self-coupling time delay fractional-order proportional-integral-derivative control strategy (ST-FOPID). The stability region of the tension control system is numerically computed and visually analyzed. Furthermore, this study improves the metaheuristic Ivy algorithm (IVYA) and proposes an improved version, the improved Ivy algorithm (IIVYA). Large-scale benchmark function experiments and performance comparisons with various algorithms, including the original Ivy algorithm, demonstrate significant improvements in convergence speed, global search capability, and avoidance of local optima. As a result, we propose a self-coupling time delay fractional-order proportional-integral-derivative control strategy (IIVYA-ST-FOPID) based on the improved Ivy algorithm, which is successfully applied to industrial systems such as simple mathematical model, DC motor speed control system, servo tension control system, and hybrid multi-area power system. Comparative analyses with various existing control strategies confirm the superior performance and broad applicability of the proposed method. Finally, both numerical simulations and experimental results show that the tension control system has excellent dynamic performance and stability under the proposed control strategy, with a significant performance improvement. Tension fluctuations can be controlled within 5% in slow winding conditions, while in fast winding scenarios, they remain around 6%. Moreover, the proposed control strategy reduces the settling time by more than 0.1 s, demonstrating the excellent performance of the proposed control strategy in complex industrial control systems.

In this paper, a formation controller for propeller-driven car-like robots is developed, which is subject to input amplitude, rate saturation, and the jointly connected topology. First, the model of the propeller-driven car-like robot is established, where actuator dynamics, amplitude, and rate saturation are considered. Second, the Gauss integration function is used to approximate the input saturation. Rate saturation will be converted into command input saturation and will be achieved by an auxiliary system. Third, the formation controller is developed based on the backstepping control method, where the adaptive robust controller and neural networks are combined to deal with unmodeled dynamics and external disturbances. According to the Lyapunov stability theory, it is proved that the propeller-driven multirobot system will be stable under the developed controller, while signals in the closed-loop system are ultimately uniformly bounded. Finally, simulation and experiment results verify the effectiveness of the proposed formation control scheme.

The C-arm, a critical device in minimally invasive surgery, captures X-ray images during operations. Positioning and orienting the C-arm, which is typically heavy, is challenging when making precise adjustments. Its nonholonomic wheel structure restricts immediate motion in any direction, complicating user control. Maneuvering the C-arm with a steering lever and pushing and pulling are inconvenient and provide imprecise movement control. This adjustment method necessitates repetitive repositioning and additional X-ray imaging, leading to delays in surgery and increased radiation exposure. This study presents a holonomic mobile platform that utilizes active casters to move a C-arm easily and accurately. This is a novel approach that has not been explored in any other study. The proposed mobile platform can automatically return a C-arm accurately to a predetermined position and orientation. In addition, a power-assisted hand-guiding function was implemented, enabling intuitive and convenient operation of the heavy C-arm and making it feel light to the user while handling and moving. The design process of the prototype is presented, and its performance is verified through experiments. In the experimental evaluations for automatic repositioning, the linear position error during automatic repositioning was found to be an average of 0.05 mm, with an orientation error of $��\; ��0.003�\; �6^{\circ }��$. The standard deviations for these errors were 0.02 mm and $��\; ��0.002�\; �8^{\circ }��$, respectively, demonstrating high reproducibility and performance.

Early identification of root disease in crops is a highly significant but challenging activity because the disease is likely to be manifested late and is also hidden in the ground. The traditional techniques of detection are slow and intrusive and lack the precision needed to act in a timely manner in precision agriculture. To overcome this, the proposed work presents a new deep learning framework called SiaRDFNet, which involves the addition of a Siamese network with a ResNet and DenseNet backbone to enhance the accuracy of root disease classification. The data set of this study was 244 high-resolution root images (126 healthy and 118 diseased samples) that were augmented by the methods of rotation, flipping, cropping, and zooming to create diversity. The framework takes the form of wavelet-based denoising, multi-scale histogram equalization, and adaptive spatial filtering, and the root segmentation when using FusionSegNet++ as the framework that introduces CBAM attention to find the exact localization of features. The segmented roots are then fed into a Siamese ResNetDenseNet fusion backbone, which allows comparative learning to be done of the healthy and diseased samples. To achieve effective disease discrimination, we use joint embedding optimization consisting of contrastive and triplet loss functions that generate a disease-sensitive and discriminating feature space. This special combination of segmentation, Siamese comparative learning, and dual embedding loss functions predetermines the originality of our approach. The model proposed showed an accuracy of 99.12%, precision of 98.02%, sensitivity of 99.16%, specificity of 99.46%, F1-score of 99.22%, and MCC of 99.15%; therefore, demonstrating that it is an effective, scalable, accurate, noninvasive solution to early root disease detection in smart agriculture.