Journal of Field Robotics最新文献

The intelligent assembly system for railway wagon bolster springs needs to realize the positioning and grabbing of bolster springs, and also has high requirements for grabbing efficiency. To solve the problem of low efficiency of traditional visual servo positioning methods, an image visual servo (IBVS) control method based on depth online detection is proposed to improve the efficiency of maintenance operations. Based on MobileNetv3 network architecture and ECA attention mechanism, a lightweight object detection ME-YOLO model is proposed to improve the real-time positioning efficiency of bolster springs. The training results show that compared with the original YOLOv5s model, the detection accuracy of ME-YOLO is slightly reduced, but the model size is reduced by 81% and the detection speed is increased by 1.7 times. Taking advantage of the real-time detection advantages of the depth camera, a visual servo control method based on depth online detection is proposed to speed up the convergence of the IBVS system. A bolster spring grasping robot experimental platform was used to conduct a visual servo bolster spring positioning comparison test. The results show that the proposed ME-YOLO detection model can meet the grabbing needs of the bolster spring assembly robot system based on IBVS, while reducing the system convergence times by about 35%. The proposed IBVS method based on deep online detection can also further improve system operation efficiency by 7%.

Transparent liquid volume estimation is crucial for robot manipulation tasks, such as pouring. However, estimating the volume of transparent liquids is a challenging problem. Most existing methods primarily focus on data collection in the real world, and the sensors are fixed to the robot body for liquid volume estimation. These approaches limit both the timeliness of the research process and the flexibility of perception. In this paper, we present SimLiquid20k, a high-fidelity synthetic data set for liquid volume estimation, and propose a YOLO-based multi-task network trained on fully synthetic data for estimating the volume of transparent liquids. Extensive experiments demonstrate that our method can effectively transfer from simulation to the real world. In scenarios involving changes in background, viewpoint, and container variations, our approach achieves an average error of 5% in real-world volume estimation. In addition, our work conducts two application experiments integrating with GPT-4, showcasing the potential of our method in service robotics. The accompanying videos and supporting Information are available at https://simliquid.github.io/.

To achieve efficient detection of ripe tomatoes in unstructured environments, this paper proposed an improved YOLOv7 rapid detection network model for ripe tomatoes. Firstly, the original YOLOv7 backbone network's CSP-Darknet53 structure was replaced by the FasterNet network structure to enhance model detection efficiency and reduce the parameters of the model. Secondly, the Global Attention Mechanism (GAM) was introduced to improve the tomato feature expression ability with a small increase in model parameters. Next, a Diverse Branch Block (DBB) module was integrated into the ELAN module in the head structure to improve the model's inference efficiency. Finally, the batch normalization layer γ was selected as the parameter of the sparsity factor in the algorithm. The L₁ regularization term was used to train the original model for sparsity, and the slim pruning algorithm was used for global channel pruning to compress the model size. The pruned model was retrained through model fine-tuning to adjust the detection accuracy to near the level before pruning. The experimental results show that the improved model has a mean average precision of 96.49%, which is basically unchanged compared to the original model. However, the model parameter count, the computation, and the model size were reduced by 52.16%, 56.84%, and 36.95%, respectively, resulting in a 32.09% increase in the recognition frame rate. Compared to similar object detection models, such as SSD, YOLOv3, YOLOv4, YOLOv5s, YOLOX, and YOLOv8, the Improved-YOLOv7 model reduced the parameter by 4.44% to 89.05%, computational complexity by 30.37% to 91.18%, and model size by 26.43% to 72.16%. This paper provided technical support for the recognition of ripe tomatoes in unstructured environments.

A better capacity for moving in restricted spaces, such as pipeline inspection and post-disaster rescue, is a key point in soft locomotion robots. Therefore, the more compact design of robots attracts great attention, including the excellent deformation ratio of the robots’ structure and the integration with the actuator. However, those pose a new challenge, that is, it is difficult to coordinate the deformation ratio between the actuator and the structure. In this study, inspired by the muscle-skeleton tension structure, a novel lever-type layout of the actuator is proposed to solve this issue. Firstly, the hollow Miura-origami structure with excellent deformation ability is adopted as the skeleton of the robot. Then, theoretical and experimental results show that the deformation amplification of the actuator can be realized with the lever-type layout, i.e., the 25% deformation ratio of the SMA spring actuator can achieve about 60% deformation ratio of the Miura-origami structure. Finally, based on the equivalent dynamic model and the non-dominated sorting genetic algorithm II (NSGA-II), the operation scenario of this new layout is demonstrated through numerical simulation. The results verify the feasibility of the new design from a simulation perspective and lay the foundation for the subsequent actuation design and control of the Miura-origami earthworm-like soft robot.

Currently, banana post-harvesting operations generally rely on manual labor, and the farming population is aging seriously, especially in hilly and mountain areas. To address these issues, the study on key technologies for mechanized banana de-handing based on an automatic feeding system was carried out, and the finite element analysis (FEA) and kinematic characteristic analysis on the de-handing cutters were conducted. The static analysis of de-handing cutters showed that the maximum stress is much smaller than the ultimate stress. Therefore, there will be no permanent deformation and no damages to de-handing cutters during the banana de-handing process. Based on the findings from FEA and kinematic characteristic analysis, the experiments to assess the self-adaptive profiling performance and de-handing performance were carried out by using the banana de-handing mechanism. The results of self-adapting profiling experiment showed that the actual profiling accuracies of the ring de-handing cutters on cylinders with diameters of 60 mm, 70 mm, 80 mm, and 90 mm were 80.80%, 80.99%, 80.90%, and 82.61%, respectively. The deviations from the corresponding theoretical profiling accuracies were 3.73%, 3.54%, 3.63%, and 1.92%, respectively. To evaluate the de-handing mechanism, the de-handing success rate, incision quality, and self-adaptive profiling effect were selected as indicators. From the de-handing experiments, it was found that the average de-handing success rate was 77.63%, the average grade of incision quality was 7.61, and the average grade of self-adaptive profiling effect was 7.70. The results of the simulation analysis and experimental study showed that the de-handing mechanism has a great stability and reliability and can meet the banana de-handing needs in the field. Compared with the de-handing cutters reported previously, the cutter proposed in this study has better profiling and enveloping capabilities for the diameter of the banana stalk, and for the first time, completely realizes the adaptive profiling of the entire stalk by the mechanical cutter, which has significant practical value for de-handing banana fruits from stalks of different diameters.

In this paper, a wrist-inspired joint with an integrated drive system is designed to realize pitch and yaw simultaneously. First, a wrist-inspired joint with a modular design using an integrated drive system is designed, which for the first time offers the flexibility of soft joints and the high rigidity of discrete joints. On the basis of this, the multilevel mapping between servos, wrist-inspired joints, and end-effectors is first analyzed. Furthermore, wrist-inspired joints are applied in manipulators and grippers. Additionally, the kinematic model of wrist-inspired joint manipulators is established based on the vector product method, and the damped least squares method is used to solve the inverse kinematics. Finally, some groups of experiments are conducted on a self-built experiment platform. Experimental results of two applications, including a manipulator and a gripper, verify the flexibility and maneuverability of designed wrist-inspired joints.

Autonomous underwater vehicles (AUVs) are specialized robots used to accomplish important field operations such as inspection of oil and gas pipelines, marine wildlife monitoring, imaging of river and sea beds, nondestructive testing of ship hulls, and so on. Before the start of an AUV mission, its navigation system, which is generally comprised of a doppler velocity log (DVL)/pressure sensor (PS)-aided inertial navigation system (INS) needs to be initialized. After a brief coarse stage of initialization, the AUV attitude is generally refined (as well as some inertial measurement unit (IMU)/aiding sensor systematic error parameters are corrected for) in a Kalman filter (KF)-based estimation process known as fine alignment, which is usually performed in open sea conditions. When the latter is conducted before the submerged phase of the AUV, a Global Navigation Satellite System (GNSS) receiver may provide additional aiding information to the refinement process. As the excitation of the degrees of freedom of the AUV is known to interfere with the performance of the KF fine alignment, this study exploits Baram and Kailath's concept of estimability to assess what kind of deliberate AUV maneuver is able to deliver the best estimation results. As the main contribution, we show that among the tested AUV motion profiles, the lawn mower is the maneuver that, except for the IMU/DVL misalignment around the AUV longitudinal axis, decreases the estimation uncertainties of all remaining INS/GNSS/DVL/PS fine alignment states. Results from simulated and experimental tests confirm the adequacy of the outlined verifications.

Labor shortages have become one of the primary challenges constraining the sustainable development of the fruit industry. The adoption of traditional multi-axis industrial-grade robotic arms for fruit picking has been limited due to issues related to efficiency and cost. This paper presents a lightweight PnP-P-R-P robotic arm that features a large workspace and an active leveling function, making it suitable for harvesting fruits such as apples and citrus. First, we establish the kinematic equations of the robotic arm and solve for the Jacobian condition number and manipulability index, showing that the workspace is free of singular points, thereby ensuring smooth operation. Next, we develop a dynamic model to analyze the performance of each joint under extreme working conditions. To adapt to practical operating environments, we simplify the forward and inverse kinematics calculations by utilizing planar spatial motion and propose a three-joint redundancy strategy for obstacle avoidance. Simulations and experimental tests reveal that the robotic arm has a vertical reach of 2.2 m and a depth of 1.3 m, with a continuous operation repeatability precision of ±5 mm when carrying a 2 kg end-effector payload. These results indicate that the robotic arm is well-suited for fruit-picking operations in both structured and unstructured orchard environments.

This paper proposes a novel 7-DOF tendon-like-driven redundant robot (TDR7) based on a weighted inverse kinematics (IK) optimization algorithm and a deep learning fine-tuning model. The robot features a modular design that enables highly flexible movements of the shoulder, elbow, and wrist joints. Its kinematic model is established using the Denavit-Hartenberg (D-H) parameter method. To address the complexity of solving IK for 7-DOF redundant robots, a weighted gradient projection method specialized for TDR7 (SWGPM-TDR7) is introduced. This algorithm integrates joint constraints, singularity avoidance, and minimum energy consumption into a multi-objective optimization framework, significantly improving joint motion continuity and trajectory planning efficiency while maintaining solution accuracy. To further accommodate complex trajectory planning requirements, a deep learning fine-tuning model (RWKV-TDR7) that combines recurrent networks with self-attention mechanisms is introduced. Through fine-tuning, RWKV-TDR7 achieves efficient trajectory fitting for TDR7, supports long-sequence outputs, and reduces computational complexity. Simulation and experimental validations demonstrate that the robot exhibits excellent performance in forward kinematics, inverse kinematics, and trajectory tracking in terms of accuracy, stability, and continuity. This work provides an effective solution for the design of high-performance robotic systems in medical and industrial applications.