SOTA Snake robots attract wide attention due to their modular bodies and multi-freedom movement abilities. However, the high production costs and intricate manufacturing procedures hinder the rapid mass production and deployment of the snake robot. Additionally, these factors limit its consumer-grade popularization and educational applications. To handle the above issues, this work proposes a low-cost lightweight intelligent modular snake robot, i.e., EdgeSnake, with a simple modular mechanical structure and flexible manufacturing technique. Specifically, each module of the snake robot is designed to minimize unnecessary volume, thereby reducing material usage and achieving a lightweight body structure. Moreover, an innovative rapid 3D printing and assembly approach is developed to efficiently realize mass production and development of specific snake robots, adapting to varying task demands and terrain features. Furthermore, EdgeSnake is also equipped with an RGB camera and onboard processor to achieve environmental perception. A comprehensive simulation control system is developed for EdgeSnake based on the ROSGazebo platform, thereby facilitating its instructional employment. Extensive experiments have been conducted on the selfbuilt simulation control system to verify the superior obstacle detouring and traversing capabilities of EdgeSnake.
Vision-and-language Navigation (VLN) is a challenging problem that requires agents to follow natural language instructions in a photo-realistic environment. The alignment between visual object information and instruction object information is critical for the navigational capabilities of intelligent agents. However, most reinforcement learning policies primarily focus on the agent’s distance change to the target viewpoint as the direct reward after taking an action, with object information playing a minor role in classical reinforcement learning for VLN. To address this limitation, we construct a new reward shaping that incorporates both the changes in the agent’s distance to the target and the progress made in navigating according to the given instruction. To capture the navigation progress, we propose an object alignment method that aligns the visual object information observed by the agent with the object information specified in the instructions. By leveraging the object’s position within the navigation instruction, we estimate the agent’s approximate progress during navigation. Experimental results demonstrate the effectiveness of our approach in reducing the navigation error (NE) and achieving high performance in terms of the success rate weighted by path length (SPL). Our method significantly enhances the agent’s ability to accurately follow natural language instructions to reach the intended destination, while also exhibiting improved generalization in unseen environments.
Intelligent machine fault diagnosis methods that leverage machine learning techniques have received widespread attention owing to their proven efficacy in enhancing production efficiency and quality as well as in reducing production costs. However, given the scarcity of rolling bearing failure data, traditional neural network training demonstrates weak noise immunity and limited network generalization capabilities. To address these issues, this study proposes a rolling bearing fault diagnosis method that combines a dual-channel hybrid domain neural network with transfer learning. The innovatively designed MDRSBU-MA module and BiGRU hybrid channel demonstrate enhanced fault feature extraction and anti-noise capabilities across domains, thus obviating the need for denoising algorithms. Furthermore, a diverse array of transfer tasks was designed for various bearing fault datasets and under differing working conditions. Experimental results suggest that the model retains robust fault diagnosis capability, particularly in terms of resistance to noise interference, small-sample cross-working condition domain, and cross-device domain transfer tasks.
Bioimpedance provides vital information for a variety of clinical application. Most current bioimpedance device are confined in their large volume, high cost and single detectable parameter. This study aimed to design a novel four-channel, multi-parameter bioimpedance acquisition device with lower volume and wearable design. This new device was designed based on an AD5933 impedance analyzer chip, enabling real-time acquisition of bio-impedance values, bio-impedance phase angle, and bio-impedance ratio through a small extension circuit for impedance acquisition and an analog switching circuit based on ADG888. Results showed that the sampling rate of the novel device could reach up to 100 Hz with a resolution of 0.01 Omega and errors within +/- 1%. Application of this novel device in bioelectrical defibrillation showed that the impedance information could successfully guide the onset and energy outputs of the defibrillation. This device may provide a novel avenue for wearable bioimpedance measurement and have a potential application in bioelectrical defibrillation.
This paper extends the Generalized UdwadiaKalaba (GUK) method to the field of control, and a novel robust control based on GUK method is proposed. The robust control can effectively handle uncertain mechanical systems with both equality and inequality constraints, in which these constraints do not affect each other. The control consists of three parts, the first part is the nominal control based on GUK equation, the second part is to compensate for the incompatibility of initial conditions, and the third part is to compensate for the uncertainty in the mechanical model. The stability analysis of the robust control is verified by applying the Lyapunov minimax method. Subsequently, through a comparative analysis of simulation outcomes involving an autonomous vehicle, this paper confirms the superiority and effectiveness of the novel robust control.
This paper studies the trajectory tracking problem for uncertain robotic manipulators with safety and performance constraints. Firstly, the safety and performance constraints can be integrated as one by choosing appropriate boundary values of performance constraints. Secondly, a state conversion function is presented to transform a constrained system into an equivalent unconstrained one and an equivalent transformed system is yielded. Thirdly, an adaptive robust controller comprised of a model-based control term, a proportional feedback control term, and a robust control term is developed to ensure that the tracking error of the transformed state is bounded, which means that the safety and performance constraints can be achieved simultaneously. Finally, the effectiveness of the proposed control strategy is validated by the numerical simulation.
In this study, a dexterous manipulation control scheme is presented, designed to facilitate the collaborative completion of various tasks using a bionic prosthesis and the healthy upper limb of a human. First, different “attractive region in environment” (ARIE) is established for specific bimanual tasks. This approach helps eliminate system uncertainty through state-independent inputs. Then a general human-robot interaction (HRI) control framework is presented, which relies on the virtual ARIE and the environment-constrained energy function. This framework is designed for collaborative tasks involving the healthy limb and prosthesis within the same workspace. Finally, a stable and unified controller consists of the environment constrained energy function and an adaptive robust method is designed for various prosthetic tasks. This controller is tailored for diverse prosthetic tasks, seamlessly integrating various task requirements and modes into a cohesive prosthetic control strategy. The proposed manipulation scheme for cooperating between the prosthesis and the healthy limb is shown to be effective by the experiment results.
In this paper, we address the problem of rapid generation of trajectories with time coordination in unmanned aerial vehicle (UAV) formations in complex urban environments. Aiming at the phenomenon that large-scale UAV system trajectory planning is computationally burdensome and fails to generate high time-optimal cooperative trajectories, we propose a time cooperative trajectory planning algorithm based on distributed model predictive control (DMPC). Firstly, a multi-UAV dynamics model based on predictive horizon is established, and a rolling iterative optimization strategy is proposed to ensure the real-time and rapidity of trajectory planning. Furthermore, the time cooperative strategy is proposed so that the UAV formation can arrive at the target position and complete the formation flight mission at the same time to ensure the feasibility of macroscopic task allocation. The simulation results show that the UAV can reach the target location smoothly in the complex environment with random assignment of obstacles, and there is no collision during the flight process, which verifies the reliability of the algorithm.
Motion planning for multi-robot cooperation in dynamic environments is challenging due to the high-dimensional nature of multi-robot systems and the presence of unpredictable obstacles. Existing motion planning algorithms often struggle to balance between motion optimality and real-time responsiveness. This paper proposes a hierarchical real-time motion planning framework to address the above challenges. We decompose the motion planning problem into two layers: a high-level centralized global planning layer and a lower-level decentralized local planning layer. The global planner is designed to perform coordinated motion optimization for long-horizon task execution trajectories of multiple robots. Concurrently, each robot is equipped with a local planner, which is capable of rapidly generating reactive motion to avoid dynamic obstacles. Simulations and experiments are conducted to validate the effectiveness of the proposed framework. The results consistently demonstrate that the hierarchical framework is capable of facilitating both efficient and safe multi-robot manipulation in dynamic environments.
In this paper, we present a novel type of supernumerary robotic limb(SRL) for the collaborative robot manipulator. It consists of a 7-degree-of-freedom (DOF) manipulator, a dexterous hand, a 5-DOF serial manipulator, and ring-type mechanisms, thus it could enhance remote flexibility and maneuverability, expand the workspace and complete complex tasks that require the cooperation of dual robotic arms. The detailed design, workspace, and kinematics of the prototype are discussed, and the experiment result by simulating the process of peeling a banana shows that the system proposed could further improve the working space and flexibility of the traditional manipulator.
The intricate multi-segment structure of human foot, which enables versatile and adaptable movement, has attracted attention to the development of biomimetic feet. However, directly imitating biological multi-segment foot structures can pose challenges in both mechanical and control systems due to their inherent complexity. This study aims to investigate the synergy and workload distribution of ankle-foot complex joints during walking to simplify the design of biomimetic multi-segment feet. We collected motion data of ankle-foot complex joints across various ground slopes (-10 degrees, -5 degrees, 0 degrees, 5 degrees, 10 degrees). We conducted hierarchical clustering analysis, focusing on seven joints: ankle, subtalar, talonavicular, calcaneocuboid, medial and lateral tarsometatarsal, and metatarsophalangeal joints, to elucidate their synergistic patterns. Based on these joint synergies, we propose an underactuated design concept for biomimetic multi-segment feet. This approach utilizes the ankle joint as the actuation source, with other joints interconnected to it. Furthermore, by analyzing the distribution of workload among joints throughout the gait cycle, we adjusted the configurations of certain joints. Specifically, we considered the calcaneocuboid, medial, and lateral tarsometatarsal joints, which exhibit roughly equivalent positive and negative work, as elastic joints, and the metatarsophalangeal joints, primarily involved in negative work, as damping joints. Simulation results confirm that our multi-segment foot design enables multidimensional humanoid motion using a single actuation source.
Multi-dimensional force sensor is a key technology to realize autonomous robot operation. In this paper, an electromagnetic multidimensional force sensor based on Hall effect is proposed. To address the issue of low accuracy in force characterization using traditional Hall effect sensors, a mechanical structure was designed to characterize the magnitude of the force by measuring the change in distance between the magnet and the Hall element. In addition, in order to enhance the number of loading/unloading cycles, an elastic film was fabricated using silica gel to realize the rebound reset effect. A specialized experimental platform was constructed to evaluate the performance of the designed sensor. The experimental results demonstrate that the sensor is capable of detecting both normal and tangential forces, as well as identifying the direction of the tangential force. This study will help to design better multidimensional force sensors.
The key to realize the application of robots in real world is to design intelligent robots with certain autonomous skill learning ability. Reinforcement learning is a feasible solution. However, two important challenges limit the application of RL methods in robotics, including the difficulty of human-designed reward as well as long training time. Therefore, we study hybrid RL methods, which use human knowledge to assist agent learning. First, we propose a reward learning method based on human preference model to realize robot skill learning, which has better robustness and convergence than the traditional RL method with human-designed reward. Then, we combine it with Episode-Fuzzy-COACH, our previous work, to build a hybrid RL method based on human preference and advice. In this method, preference model is used to infer reward function and human advice is used to speed up the policy learning process. It realizes efficient robot skill learning without human-designed reward function. And it is proven the learning efficiency of this method is 73.3% higher than that of the reward learning method that only uses preference model.
Ultrasound (US) imaging is an important, non-invasive method for evaluating the carotid artery (CA), which plays a critical role in early screening cardiovascular and cerebrovascular conditions. However, it requires specialized training for sonographers, which is a challenge in primary healthcare. Robotic devices now enable semi-automatic or fully automatic US scans, reducing the need for skilled operators. This paper presents an innovative, economically viable robotic system specifically engineered for conducting US examinations of the human CA. This system is distinguished by its advanced human-machine interaction capabilities with artificial intelligence assistance, which facilitate semi-automatic scanning processes in primary healthcare settings. Building upon this, the study introduces a semi-automatic scanning approach designed to control the device’s operations, aiming at capturing highquality images of the key structures of CA. Employing the images obtained through this method, the research proposes a segmentation strategy, leveraging fine-tuning adjustments on the LiteMedSAM model to achieve semantic segmentation of the CA in the images. The outcome is a nuanced diagnostic tool capable of identifying and evaluating carotid stenosis. The mentioned device and methods are tested on human subjects and the results indicate success on detecting critical vascular structures and positive feedback on human-machine interaction from operators and patients.
This paper presents a novel modeling approach of an articulated-vehicle system comprising dual multi-axle trailers which does not belong to conventional trailer systems such as multi-steering n-trailer and multi-steering general n trailer systems. In the new model, some existing mechanical elements are replaced with virtual counterparts, and furthermore, virtual mechanical elements are incorporated, facilitating the conversion of the kinematical equation of the articulated-vehicle system into a chained form. This paper also presents a control law, derived from the chained form, for enabling four manipulation points defined on the articulated-vehicle system to follow a single desired free-form curve path. The validity of the modeling approach and the control law is verified by experiments.
Exoskeleton robots are a useful technique for hand rehabilitation. Most current exoskeleton robots have low active degrees of freedoms (DOFs) and confined range of motion (ROM), which may constrain the hand dexterity and motion area for rehabilitation. This study aimed to develop a novel exoskeleton robot with more active DOFs and larger ROM for hand rehabilitation. The novel exoskeleton robot was designed with 22 DOFs. The mechanical structure of the exoskeleton was optimized by kinematic modeling and optimization problem construction. The ROM was evaluated using extraction of the structure optimized robot motion space envelope. The effect of the novel exoskeleton was examined by an experiment using data of brain tissue oxy-hemoglobin (HbO). This study collected brain tissue HbO concentrations without training and with the assistance of the exoskeleton robot for comparison. Results showed that the ROM of the exoskeleton was improved from 169 cm(3) to 207 cm(3). The mean and maximum concentrations of brain tissue HbO during hand motion driven by the exoskeleton were 0.0429 (mmol/L) *mm and 0.0921 (mmol/L) *mm, respectively. By contrast, the mean and maximum without exoskeleton-assistant motion were -0.0541 (mmol/L) *mm and 0.0138 (mmol/L) *mm, respectively. These results suggested that the brain tissue HbO during exoskeleton-assisted training was notably elevated compared to non-training conditions. This is a preliminary study showing the effectiveness of the novel exoskeleton robot.
Adverse weather conditions significantly impact the reliability of perception systems in autonomous driving. To address this issue, we designed an iterative point cloud denoising method that iteratively adds displacement to help the point cloud converge to the ground truth. We constructed a foggy weather dataset based on KITTI to simulate foggy driving environments. Experimental results demonstrate that our method outperforms state-of-the-art methods and significantly improves the quality of denoised point clouds. Furthermore, we used our method as a preprocessing step for multiple point cloud registration modules, and the results showed that our method exhibits good generalization across different registration algorithms. It is well-suited as a preprocessing module embedded in downstream point cloud processing tasks.
LiDAR plays an important role in autonomous driving since it can provide accurate 3D scene structure information by generating point cloud data. A key challenge in 3D object detection comes from the sparse distribution of 3D points caused by occlusion, the distant target or the limitation of LiDAR attributes, resulting in high rates of False Negatives. In this paper, we introduce an anchor-based transformer for temporal LiDAR 3D object detection, which can reduce the missed detections by utilizing temporal prior knowledge. Specifically, the proposed method first utilizes the sparse convolution to process the point voxels to generate downsample BEV feature representations, which are then fed into our anchor based transformer to establish long-range attention. Then, the proposal anchors from previous and current frames are aggregated and encoded as the queries to extract all candidate targets features in the deformable attention module. This process can enhance the target features weakened by sparse point distribution, thereby elevating the overall detection performance. On the proposed Port dataset, our method demonstrates a significant improvement over existing state-of-the-art methods, thereby verifying the effectiveness of our proposed approach.
Ventilation pipes are an important way of supplying fresh air to a room. However, there are cases where corrosion and dirt accumulation require inspection of the inside of the pipes. The development of pipeline inspection robots is an effective solution, but for the inspection task of narrow metal ventilation pipes, the existing pipeline robots are limited in size, kinematic capability and perception. In this study, a miniaturised pipeline robot is proposed and developed with a dual articulated arm crawler as the basic kinematic mechanism, which has the ability to cross obstacles, climb slopes and adapt to diameter changes in narrow pipelines. The pipeline robot sensing system is designed in a modular way, using the vanishing point and camera projection model to estimate the position of the robot in a straight metal pipe environment, and analysing the characteristics of the ventilation pipe system to construct a scalable pipeline topology map based on a weighted undirected graph. Finally, a ventilation pipe scene is constructed to conduct experiments on the pipeline robot. The results show that the designed pipeline robot can move flexibly in a narrow metal pipe environment, complete localization within the pipe, and construct pipe maps.
This work addresses the single beacon cooperative localization for autonomous underwater vehicles. The unreliable acoustic communication channel is used to obtain the relative range measurements that suffer from outliers, which may lead to casual failure with divergence. To tackle this issue, an adaptive moving horizon estimation framework is proposed for robust localization. Based on the single beacon cooperative localization model and moving horizon estimation, the fixed-dimensional optimization problem is formulated in terms of two adaptive regularizations, the observability one and the indirect velocity one. The uncertainty of position represented by covariance is added to observability regularization, it will be adaptively modified according to the observability of state to adjust the weight of objective function. Physical constraints are considered in the indirect velocity regularization through available proprioceptive measurements. To obtain the covariance and realize fast convergence, the real-time estimator is designed to fastly calculate the initial suboptimal estimation of the optimization problem. Numerical simulations demonstrate that the localization performance of the proposed adaptive estimation is more robust.
