Journal of Field Robotics最新文献

Establishing common knowledge about environmental conditions, task objectives, and coordination rules is crucial for improving the collaborative efficiency of swarm robots. In complex scenarios, relying on a centralized facility to maintain this knowledge is impractical, necessitating a decentralized approach. Blockchain technology offers a promising solution for decentralization and can tolerate some degree of malicious or malfunctioning entities. However, widely used blockchain approaches, such as those employed in Ethereum and relying on proof-of-work (PoW) or proof-of-authority (PoA), demand significant computational resources, rendering them impractical for swarm robotics applications. This paper introduces PTEE-BFT, a novel parallel Byzantine fault tolerance protocol leveraging the trusted platform module (TPM). PTEE-BFT employs a Unique Sequential Identifier Generator (USIG) to ensure the monotonicity, uniqueness, and order of messages, thereby reducing the number of communication phases and replicas required. This significantly enhances the efficiency and fault tolerance of the consensus process. Additionally, PTEE-BFT implements parallel processing strategies to substantially increase blockchain system throughput. Furthermore, we develop an algorithm that enables the robot swarm to recognize attacks from a specific type of malicious robot known as Byzantine robots. Our experimental analysis and performance evaluation demonstrate that PTEE-BFT achieves an optimal balance among performance, scalability, and fault tolerance, outperforming practical Byzantine fault tolerance (PBFT). Results from physical robots show that our approach significantly reduces computing overhead and accelerates consensus formation compared to baseline solutions. This represents a significant advancement in blockchain consensus mechanisms for swarm robotics.

Quadrupedal robots are highly regarded for their superior locomotion capabilities and terrain adaptability, making them competent in a wide range of applications. For autonomous navigation, they must track upper-level trajectories to reach designated locations with flexible obstacle avoidance. This is typically achieved by a planner, which generates a reference velocity, and a controller, which accurately tracks the velocity commands. This article proposes a learning-based controller for quadrupedal robots that is trained in simulation and achieves accurate and robust velocity tracking in the real world. To bridge the gap betwen simulation and reality, an analytical actuator model is introduced to simulation to simulate physical actuator dynamics. We then train a control policy in simulation using Constrained Reinforcement Learning, where symmetry and smoothness constraints are incorporated into Reinforcement Learning. The symmetry constraint promotes coordinated locomotion and consistent velocity tracking performance, while the smoothness constraint reduces jerky actions and generates stable velocity performance. The proposed control policy is zero-shot deployed on the Unitree AlienGo. Experimental results demonstrate a velocity tracking error below 0.084 m/s across the entire operational velocity range while maintaining robust locomotion on natural terrains. To further validate the controller's effectiveness, we integrate it into a pedestrian tracking framework, where it demonstrates precise trajectory following capabilities and long-term reliability.

Assessing the operating environment of a quadruped robot and analyzing the forces acting on its footpads are crucial for enhancing motion stability and enabling effective path planning, ultimately ensuring successful task completion. This study focuses on analyzing the robot's motion state and its contact process to better understand its contact dynamics and develop a comprehensive dynamic model. The model defines the relationship between footpad forces and joint angular torque, enabling both environmental assessment and force prediction. Experimental validation was conducted using the quadruped robot in various environments, confirming the model's effectiveness. By comparing the joint angular torque during contact and noncontact states, the footpad contact conditions were determined. During the contact phase, joint angular torque exceeded the noncontact torque, with greater discrepancies corresponding to higher footpad forces. These findings suggest that denser soil improves the robot's performance. The method for calculating footpad forces from torque demonstrated accuracy exceeding 90%, highlighting its precision. These results provide valuable insights for calculating operational forces, as well as for the stability assessment and trajectory planning of quadruped robots.

Visually impaired individuals (VIIs) constitute a significant portion of the global population, residing in various regions worldwide. There are current issues that include problems detecting obstacles, low landmark availability for orientation, navigation dependency on other people, and understanding visual data in the new environment. To address these issues, an advanced model called Enhanced Convoluted Elk Herd Lightweight Yolo-fastest V2 (ECEHLY-V2) is proposed. This model aims to enhance both functionality and user-friendliness in navigating environments with obstacles. The ECEHLY-V2 model derives features from detected objects by utilizing object detection and tracking mechanisms, enabling overall obstacle detection. The combination of the improved elk herd optimizer with the convoluted lightweight YOLO-fastest V2 model enhances obstacle detection and tracking by optimizing model parameters for higher accuracy, accelerating convergence during training, and guaranteeing effective resource use. With critical assessment, the suggested method has impressive performance statistics, such as accuracy at 99.8%, precision at 98.65%, recall at 98.25%, and F1-score at 97.64%, which is better than other techniques. Therefore, the ECEHLY-V2 model is a notable innovation in navigation aid technology for VIIs, providing strong obstacle detection and tracking abilities to promote their safety and mobility in real-life situations.

This paper presents a biomimetic inchworm magnetic wall-climbing robot, which meets the needs of on-site welding, grinding, inspection, and other tasks for large steel structures. The robot is innovatively proposed by observing the movement of the inchworm, which can not only achieve the full internal corner wall adaptation between the ground and the vertical wall, and between the vertical wall and the ceiling, but also achieve the full external corner wall adaptation between the vertical wall and the roof and between the ceiling and the upper vertical wall. The mechanical behavior of the robot on vertical walls, inclined walls, ceiling, and traversing between various walls is thoroughly examined. The analysis provides essential mechanical criteria to ensure the robot's secure operation under various working conditions. To further validate the structural design's robustness, we employ finite element analysis (FEA) on the critical structural components using ANSYS. At the same time, ANSYS Maxwell was used to simulate and analyze the magnetic attraction, and then the magnetic attraction curves across the internal and external angles of the robot were analyzed to ensure the safety of the robot's movement. Finally, according to the design and analysis, the prototype was developed and tested, and the test results showed that the robot met the expected functions and indicators.

Most existing vision-based simultaneous localization and mapping systems and their variants still assume that the observation is absolutely static and cannot work well in dynamic environments. In this paper, we propose a direct geometrically constrained SLAM method based on target detection and depth image segmentation, named YGDD-SLAM. The YGDD-SLAM system can work robustly, accurately, and continuously in highly dynamic environments. The method first acquires static and potential dynamic feature points in the current frame through a target detection network. Then, dynamic targets are identified by combining the geometric change relationship between static and potential dynamic feature points between adjacent frames. To improve the accuracy of the dynamic judgment, the motion probability of the potential dynamic target in the past few frames is also used for judgment. Subsequently, the dynamic object regions at the pixel level are segmented out based on the double-peak feature of the gray-scale histogram of the dynamic target region in the depth image, which ultimately achieves the accurate deletion of all dynamic features points. Meanwhile, we validate YGDD-SLAM on TUM data set and Bonn data set and prove that it significantly improves the localization accuracy and system stability in different types of dynamic environments.

This study focuses on implementing a super-twisting controller (STC) to manage autonomous rover navigation. STCs have effectively dealt with inherent uncertainties in real-world applications, making them particularly suitable for tasks such as rover navigation. The study addresses designing and implementing an STC-based controlled system tailored to rover navigation scenarios' dynamic and unpredictable nature. STC uses equivalent control schemes to model complex relationships among input variables, minimizing errors in wheel speed, steer mechanism, and 6-DOF robot arm trajectories. The implementation is validated through simulation studies, using representative scenarios to evaluate the controller's performance in demanding environments. Evaluation metrics include trajectory accuracy, obstacle avoidance, and overall system robustness. In addition, the same control action was tested on a rover that was developed by the authors, whose motion corresponds to that considered in the numerical evaluations. The results of this study are intended to provide valuable information on the application of STC for autonomous rover navigation, with implications for improving the reliability and adaptability of robotic exploration missions.

For fragile and delicate fruit in agricultural production, damageless robotic grasping is still an open problem. In this review, we aim to synthesize existing studies and provide insights on this issue. Compared with other related reviews, we focus on fragile fruit and damage problems, and cover more complete stages and technical aspects. The referenced studies were divided into two categories according to their stages, including picking and harvesting in the preharvest stage, and uploading and packaging in the postharvest stage. In this context, the systematic literature review method was adopted for literature collection and selection. The references were classified into four categories according to their technical aspects: system design, end effector, visual feedback, and tactile feedback. Additionally, recent progress in commercial products and companies was also investigated. In summary, techniques such as a dedicated end effector for fragile fruit, advanced tactile sensing, and control strategy are supposed to be further studied to achieve damageless robotic fragile fruit grasping. Furthermore, other advanced perspectives are potentially becoming future trends, such as in-hand sensing of fruit quality and dexterous manipulation for complex tasks. We hope this review can provide insights in both academia and industry to promote robotization in fruit production.

Aiming at the problem of locating the faulty bearing of robot propeller, the traditional mechanism modeling method is difficult to establish and cannot cover all unknown factors due to too many factors involved; however, models that rely on actual data are difficult to provide enough training samples due to expensive equipment and complex working conditions. Therefore, an agent model based on knowledge and data fusion is proposed to accurately capture the mapping relationship between input data and fault-bearing location. First, the mechanism model of the propeller-bearing rotor system under ideal working conditions is established, and data samples are generated. Then, a fault location agent model-building method based on incremental knowledge distillation is proposed, which integrates mechanism knowledge with actual data. Combined with the data continuously obtained in actual operation, the agent model is continuously updated and optimized to enable it to dynamically adjust under different working conditions and improve the fault location accuracy. Finally, the fault feature attribute description strategy is embedded in the agent model to make the representation of mechanism knowledge and data more consistent, so as to achieve more effective integration of the two. The experimental results show that the agent model not only significantly reduces the complexity of building the model, but also can accurately reflect the mapping relationship between data and fault-bearing location under different working conditions through only one measuring point, so as to achieve accurate diagnosis of fault-bearing location.