Journal of Field Robotics最新文献

In this paper, an adaptive tracking controller for the propeller-driven wall-climbing robot is developed, which is subject to velocity-related input saturation and velocity constraint. First, the model of the propeller-driven wall-climbing robot is established, where actuator dynamics and input saturation are considered with velocity constraints. The strategy of active gravity balance is put forward, which simplifies the modeling but leads to the problem of velocity-related input saturation. Second, the Gauss integration function is used to approximate the velocity-related input saturation. The velocity constraint would be handled by employing the barrier Lyapunov-based transformation rather than the barrier Lyapunov function (BLF) method. Thirdly, the tracking controller is developed based on the dynamic surface 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 robot system will be stable under the developed controller, while signals in the closed-loop system are ultimately uniformly bounded. Finally, simulation results show the effectiveness of the proposed tracking control scheme.

Wheeled rovers have been widely used to conduct surface exploration on the Moon or Mars. However, these rovers struggle to traverse granular steep slopes, thereby limiting potential exploration sites. Biomimetic legged robots offer superior mobility compared to wheeled rovers and are expected to enhance surface exploration capabilities. Nevertheless, current biomimetic robots exhibit limited climbing ability on steep slopes. Inspired by desert lizards that move efficiently on granular sand, this study proposes a biomimetic transition gait to enable a quadruped robot to creep from the ground onto slope surfaces. When ascending a slope, the robot elevates its trunk above the incline while ensuring that all feet maintain contact with the sand. During leg swings, the trunk remains attached to the slope to prevent slippage. Combined with active attitude adjustments, the robot can stably move from ground to slope on a Martian soil analog testbed. In experiments on level ground to 32° of Mars slope analog, the robot demonstrated a transition speed of 2.83 mm/s, thereby advancing the capability of quadruped robots to explore uneven terrain on the Moon or Mars.

Broccoli head recognition algorithms often demonstrate poor robustness under complex natural lighting conditions. Additionally, commonly used instance segmentation methods require extensive manual annotation, making their application to other broccoli datasets challenging. Improving detection accuracy and reducing annotation costs are crucial for efficient in-field broccoli head recognition. We aimed to develop a robust, transferable broccoli head detection and diameter estimation algorithm that performs accurately under complex lighting conditions and leaf occlusion. We proposed the RSDF algorithm, integrating YOLOv8n for object detection, SAM2 for segmentation, and depth-based contour denoising. Field tests were conducted under mixed lighting conditions with shading and auxiliary illumination. The proposed RSDF algorithm demonstrated high robustness, maintaining a recognition rate above 95% across different lighting conditions. Under shading conditions, the segmentation performance of SAM2 remained stable, with a coefficient of variation within 3% despite changes in auxiliary lighting intensity. Testing the RSDF algorithm on 100 field images with shaded auxiliary lighting yielded a detection accuracy of 96.9% and a head diameter estimation accuracy of 90.31%. The RSDF algorithm enhances broccoli detection in challenging environments, offering a scalable solution for automated harvesting.

The present study proposes a new design for a robot fish which is propelled by periodic vibrations of a flexible-soft coupled beam. This unique, flexible-soft coupled beam can display the first and second-order mode oscillations, which are capable of effectively mimicking the swings of the fish tail, resulting in fast swimming of the robot fish. Firstly, the nonlinear dynamic model for the coupled beam is established based on the absolute node coordinate formulation. The effect of length and stiffness parameters on natural frequency and mode shape of the coupled beam is explored to obtain the linear dynamic characteristics of the fish tail. To reach the maximum vibration amplitude, the optimal values for position, frequency, and magnitude of the applied force and the length and stiffness parameters are determined. In the following, the control system uses a communication mode to receive signals from a wireless communication module, and an inertial sensor is designed. The fuzzy PID algorithm is employed to control vibrations of the coupled beam to realize the swimming forward and turning around of the robot fish. Finally, through 3D printing and the opening mold technique, the robot fish is fabricated with an overall size of 130 × 125 × 70 mm³. Swimming experiments are performed to display the propulsion speed and force of the robot fish. It shows that the swimming speed of 1.17 BL/s can be achieved, which is much higher than most of the previously designed robot fish in BCF mode. In addition, the experiments indicate that the robot fish has an excellent swimming performance even in countercurrent, complex trajectories, and heavy load environments. The present study offers a delicate design and a precise theory of the flexible-soft coupled beam-based fish tail for fast swimming of the robot fish.

This paper presents a novel Recurrent Neural Network (RNN) controller for redundancy resolution and orientation control of the Stewart platform. The Stewart platform features six prismatic actuators, making it a six-degrees-of-freedom (6-DOF) system. When imposing three-dimensional orientation control, the platform retains a redundancy of 3-DOF, which can be utilized to achieve secondary goals. The key novelty of this study lies in the formulation of a Jacobian-free, gradient-free control strategy that directly solves a constrained nonlinear optimization problem at the angular level, thereby significantly improving computational efficiency and robustness compared with conventional controllers. Specifically, we propose the Beetle Antennae Olfactory Recurrent Neural Network (BAORNN) algorithm, a biologically inspired metaheuristic framework that bypasses the computationally intensive Jacobian inversion typically required in redundancy resolution. The orientation control problem is formulated as a constrained optimization task, incorporating an energy-efficient actuator usage objective and mechanical constraints modeled as inequalities. Theoretical stability and convergence guarantees are established for the proposed BAORNN framework, ensuring reliable operation across a wide range of configurations. To validate the approach, we developed a high-fidelity simulation environment using the Simscape Multibody library in Simulink and conducted extensive experiments across multiple time-varying reference trajectories. Quantitative performance comparisons against a state-of-the-art inverse kinematics controller demonstrate the superior accuracy, convergence speed, and constraint-handling capabilities of our method. Furthermore, we showcase a realistic application scenario by integrating the controller with a chair-mounted Stewart platform for immersive driving and flight simulations, demonstrating the potential for real-world deployment in motion simulation and training systems. In summary, this paper introduces a computationally lightweight, robust, and highly accurate RNN-based controller tailored for redundant Stewart platforms, with proven advantages over traditional Jacobian–based methods.

Real-time object recognition is a significant field of research with numerous applications, including object tracking, video surveillance, and autonomous driving. This identifies the smallest bounding boxes that encompass the objects of interest within the input images. Nevertheless, these approaches face challenges, like limited support for quantization and suboptimal trade in achieving accurate object detection. To address these issues, a novel approach called Faster region-based Convoluted Non-monopolize search You Only Look Once neural architecture Search (FCN-YOLOS) is introduced for object detection. This approach merges the advanced feature abstraction abilities of Faster R-CNN with the efficient object recognition strengths of YOLOv8, enhanced by NAS optimization. YOLOv8 is employed for its rapid and accurate real-time detection of abandoned items, while Faster R-CNN contributes sophisticated feature extraction by utilizing statistical, grid, and Histogram of Oriented Optical Flow (HOOF) features to improve object representation and classification. Additionally, NAS optimizes hyperparameters by balancing exploration and exploitation, which helps minimize the loss function, reduce overfitting, and enhance generalization. This results in exceptional real-time object detection performance within the FCN-YOLOS framework. The proposed technique has demonstrated a maximum image of approximately 99%, 96.3%, 94.9%, and 95.2% concerning brightness realization compared to existing methods for accuracy, recall, precision, and F1 score, respectively. These outcomes highlight its extensive applicability across diverse object detection contexts, rendering it a compelling option for both academic and industrial research. Overall, the proposed approach for object recognition techniques in feature extraction and hyperparameter adjustments further improves evaluation in terms of efficiency and object detection accuracy.

This article presents an algorithm for mobile robots that enables autonomous navigation in complex environments. Currently, achieving autonomous navigation for ground mobile robots in intricate and unstructured settings continues to pose significant challenges. To address issues such as dispersed sampling points, low sampling efficiency, and excessive path waypoints encountered in traditional Rapidly-Exploring Random Trees (RRT) algorithms, this paper proposes an Optimized Sampling Strategy and Artificial Potential Fields Fusion-based Informed-RRT* global path planning algorithm. Initially, sampling angles are determined based on the position of the target point, and the workspace is partitioned into regions with varying levels of importance. Subsequently, an improved artificial potential fields algorithm is integrated to further refine the resultant forces acting on the nodes. Finally, cubic spline interpolation is utilized to smooth the generated path. The proposed algorithm was validated through simulation and experimental studies conducted on simple, narrow, and complex maps. The results demonstrated significant reductions in search time, path length, and the number of path waypoints compared to conventional A*, Dijkstra, RRT, RRT*, and Informed-RRT algorithms. Additionally, the smoothness of the generated paths was notably improved. In the virtual maze experiments and real-world environment tests, the improved algorithm presented in this paper demonstrates significant advantages over five other algorithms.

Odometry plays a crucial role in autonomous tasks of field robots, providing accurate position and orientation derived from sequential sensor observations. Odometry based on Light Detection and Ranging (LiDAR) sensors has demonstrated widespread applicability in environments with rich structured features, such as urban and indoor settings. However, for unstructured environments like scrubland and rural roads, the extraction, description, and correct matching of LiDAR features between frames become challenging. Due to the lack of flat surfaces and straight lines, the existing odometry approaches, whether using hand-crafted features such as edge and planar points or learned features through networks, will face the problem of decreased positioning accuracy and potential failure. Therefore, we propose a neural LiDAR odometry based on Trans-frame Association to extract more effective features for pose estimation in unstructured environments. The Trans-frame Association module contains a fully interactive frame Transformer and a scan-aware Swin Transformer. The former applies cross-attention to features extracted from two consecutive frames, thus enhancing the accuracy and robustness of feature correspondences by considering the contextual information. The latter restricts the attention mechanism to shift along the scan lines of LiDAR, thereby leveraging the sensor's inherent higher horizontal resolution. Our Transformer has linear complexity, which guarantees the module can meet real-time requirements. Additionally, we design a Reuse Refinement Pyramid architecture to further improve the accuracy of pose estimation by reusing multiresolution features. We conducted extensive experiments on the RELLIS-3D data set and our Matian Ridge data set collected in a representative unstructured scene. The results demonstrate that our network outperforms recent learning-based LiDAR odometry methods in terms of accuracy. The code is available at https://github.com/qlsinori/FAR-LO.

In the field of high-redundancy manipulators, specifically in the rock drilling manipulator domain, fast and efficient path planning is crucial. Therefore, this paper proposes an improved algorithm, v-BI-RRT, based on the BI-RRT algorithm and oriented vector methods. In this algorithm, the nodes along one path are extended in the direction of the node coordinates of another path as the target direction. When the path collides with an obstacle, new node coordinates are generated using a random sampling method to bypass the obstacle. This approach enhances spatial search efficiency. For high-redundancy manipulators like the rock drilling manipulator, self-collision avoidance is a key component of collision-free path planning. This paper uses oriented bounding boxes (OBB) and capsules to envelope the manipulator's body. Potential self-collisions are detected in two stages: during the rapid detection phase, non-colliding pairs are quickly excluded, and during the precise detection phase, the distance between the remaining potential collision pairs is calculated using Euclidean distance to find the shortest distance. Finally, the self-collision detection algorithm is integrated into the v-BI-RRT algorithm. Simulations and experiments demonstrate that the algorithm responds quickly and performs well in avoiding collisions when applied to path planning for the rock drilling manipulator.