Human Activity Recognition (HAR) holds significant importance in health and human-machine interaction. However, recognizing actions from 2D information faces challenges like occlusion, illumination variation, cluttered backgrounds, and view invariance. These hurdles are particularly pronounced in indoor patient monitoring settings due to fluctuating lighting conditions and cluttered backgrounds, which compromise the accuracy of activity recognition systems. A new architecture named IlluminationRevive has been proposed to tackle this issue, which utilizes an encoder-decoder convolutional neural network (CNN) and image post-processing blocks to enhance the image’s visual appearance. A new dataset comprising seven indoor physical activities has been proposed, created with contributions from thirty individuals aged 20–45. A hybrid fusion architecture is proposed to classify activities, integrating motion sequence information and body joint features. The proposed classification model incorporates generated Skeleton Motion History Images (SMHIs), collected human joint motion features from video frames, and novel kinematic and geometric features within window frames as inputs. The model can extract spatial and temporal feature vectors by integrating ResNet50-ViT (Residual Network-50 layers, Vision Transformer) and CNN-BiLSTM (Convolutional Neural Network-Bidirectional Long Short-Term Memory) layers. The suggested classification model was evaluated alongside state-of-the-art models using the LNMIIT-SMD (The LNM Institute of Information Technology-Skeleton Motion Dataset) and established NTU-RGBD (Nanyang Technological University’s Red Blue Green and Depth information) dataset. The evaluation aimed to assess the effectiveness of the proposed classification model architecture. Results demonstrate the superiority of the proposed model, achieving impressive accuracies of 98.21% on real-time data, 98.45% on the proposed dataset, and 97.12% on the NTU-RGBD dataset. This high-accuracy, low-latency approach enhances robotic perception for healthcare applications, enabling service robots to perform real-time patient monitoring and assistive tasks in dynamic indoor environments.

Multi-loop coupling mechanisms (MCMs) are extensively utilized in the aerospace and aviation industries. This paper analyzes the mobility, singularity, and optimal actuation selection of a 3RR-3RRR MCM on the basis of geometric algebra (GA), where R denotes revolute joint. First, the principle of the shortest path is employed to identify the basic limbs and ascertain the type of coupling limbs. The analytical expression for the twist space and mobility characteristics of the mechanism is obtained by calculating the intersection of the limb’s twist space. The blade of limb constraint is subsequently employed to construct the singular polynomials of the mechanism. The singular configurations of the 3RR-3RRR MCM are analyzed in accordance with the properties of the outer product, resulting in the identification of two distinct types of boundary singularities. Next, the local transmission index is employed to evaluate the motion/force transmission performance of the two actuation schemes and finalize the selection of the superior actuation scheme for the mechanism. Finally, a prototype is developed to evaluate the energy loss resulting from the two actuation schemes, which verifies the correctness of the actuation selection scheme.

This paper proposes a generalized method for designing tendon-driven serial-chained manipulators with an arbitrary number of tendon redundancy. First, a special class of tendon-driven structures is defined by introducing the controllable block triangular form (CBTF) of a null space matrix and its complementary CBTF of a structure matrix, satisfying physical constraints related to the minimal connection of tendons and to the placement of actuators. Then it is shown that any general design of tendon-driven serial manipulators can be reduced to the design of such a special class of tendon-driven structures. Two associated design problems are derived and solved. The first design problem is about finding a complementary CBTF structure matrix for a given CBTF null space matrix using algebraic relations, whereas the second one seeks the both matrices that optimize the wanted structural characteristics based on the result of the first design problem. Numerical design examples are provided to show the validity of the proposed method.

Underwater robots conducting inspections require autonomous obstacle avoidance capabilities to ensure safe operations. Training methods based on reinforcement learning (RL) can effectively develop autonomous obstacle avoidance strategies for underwater robots; however, training in real environments carries significant risks and can easily result in robot damage. This paper proposes a Sim-to-Real pipeline for RL-based training of autonomous obstacle avoidance in underwater robots, addressing the challenges associated with training and deploying RL methods for obstacle avoidance in this context. We establish a simulation model and environment for underwater robot training based on the mathematical model of the robot, comprehensively reducing the gap between simulation and reality in terms of system inputs, modeling, and outputs. Experimental results demonstrate that our high-fidelity simulation system effectively facilitates the training of autonomous obstacle avoidance algorithms, achieving a 94% success rate in obstacle avoidance and collision-free operation exceeding 5000 steps in virtual environments. Directly transferring the trained strategy to a real robot successfully performed obstacle avoidance experiments in a pool, validating the effectiveness of our method for autonomous strategy training and sim-to-real transfer in underwater robots.

The autonomous navigation and obstacle avoidance capabilities of autonomous underwater vehicles (AUVs) are essential for ensuring their safe navigation and long-term, efficient operation. However, the complexity of the marine environment poses significant challenges to safe and effective obstacle avoidance. To address this issue, this study proposes an AUV obstacle avoidance control algorithm based on offline reinforcement learning. This method adopts the Conservative Q-learning (CQL) algorithm, which is based on the Soft Actor-Critic (SAC) framework. It learns from obtained historical obstacle avoidance data and ultimately achieves a favorable obstacle avoidance control strategy. In this method, PID and SAC control algorithms are utilized to generate expert obstacle avoidance data to construct a diversified offline database. Additionally, based on the line-of-sight (LOS) guidance method and artificial potential field (APF) method, information regarding the distance and orientation of targets and obstacles is incorporated into the state space, and heading and obstacle avoidance reward terms are integrated into the reward function design. The algorithm successfully guides the AUV in autonomous navigation and dynamic obstacle avoidance in three-dimensional space. Furthermore, the algorithm exhibits a certain degree of anti-interference capability against uncertain disturbances and ocean currents, enhancing the safety and robustness of the AUV system. Simulation results fully demonstrate the feasibility and effectiveness of the intelligent obstacle avoidance method based on offline reinforcement learning. This study highlights the profound significance of offline reinforcement learning in enabling robust and reliable control systems for AUVs, paving the way for enhanced operational capabilities in challenging marine environments.

This paper investigates a closed-loop visual servo control scheme for controlling the position of a fully constrained cable-driven parallel robot (CDPR) designed for functional rehabilitation tasks. The control system incorporates real-time position correction using an Intel RealSense camera. Our CDPR features four cables exiting from pulleys, driven by AC servomotors, to move the moving platform (MP). The focus of this work is the development of a control scheme for a closed-loop visual servoing system utilizing depth/RGB images. The developed algorithm uses this data to determine the actual Cartesian position of the MP, which is then compared to the desired position to calculate the required Cartesian displacement. This displacement is fed into the inverse kinematic model to generate the servomotor commands. Three types of trajectories (circular, square, and triangular) are used to test the controller’s compliance with its position. Compared to the open-loop control of the robot, the new control system increases positional accuracy and effectively handles cable behavior, various perturbations, and modeling errors. The obtained results showed significant improvements in control performance, notably reduced root mean square error and maximal error in terms of position.

Collaborative robotics in manufacturing introduces a new era of seamless human–robot collaboration (HRC), enhancing production line efficiency and adaptability. However, guaranteeing safe interaction while maintaining performance objectives presents significant challenges. Integrating safety with optimal robot performance is paramount to minimize task time and ensure its completion. Our work introduces an architecture for safety in confined human–robot workspaces by integrating existing safety and productivity methods into a unified framework specifically designed for constrained environments. By employing an improved artificial potential field, we optimize paths based on length and bending energy and compare baseline algorithms like gradient descent algorithm and rapidly exploring random tree (RRT*). We propose an evaluation metric for system performance that objectively maps to the system’s safety and efficiency in diverse collaborative scenarios. Additionally, the architecture supports multimodal interaction, including gesture-based inputs, for intuitive control and improved operator experience. Safety measures address static and dynamic obstacles using potential fields and safety zones, with a real-time safety evaluation module adjusting trajectories under specified constraints. A performance recovery algorithm facilitates swift resumption of high-speed operations post safety interventions. Validation includes comparing the algorithmic performance through simulations and experiments using the 6-degrees of freedom UR5 robot by universal robots to identify the most suitable algorithm. Results demonstrate an 83.87% improvement in system performance compared to ideal case scenarios, validating the effectiveness of the proposed architecture, evaluation metric, and multimodal interaction in enhancing safety and productivity.

This paper presents the design and implementation of Jaeger UTFPR, an open-source, low-cost, remote-controlled tiny humanoid robot measuring just 12 cm in height. Developed with a focus on accessibility and affordability, the robot integrates 3D-printed components, cost-effective electronics, embedded systems, and wireless communication to provide real-time audio and video feedback through a virtual reality (VR) interface. Operators control Jaeger UTFPR using a VR headset and motion controllers, enabling immersive telepresence and direct manipulation of the robot’s movements. With a total cost of just a few tens of dollars, this innovative solution offers broad applications in education, entertainment, research, and remote inspection, serving as an accessible platform for robotics enthusiasts and developers. Experimental evaluations demonstrate the system’s effectiveness in balancing performance and cost, validating its potential as a tool for immersive robotics experiences.

Humanoid robots are highly redundant, and finding whole-body optimal trajectories for various tasks is very complex. This paper proposes a method to find an energy-optimal, dynamically balanced, and collision-free trajectory of the 20 degrees of freedom humanoid robot in pick and place application. The task of pick and place is divided into three subtasks using the Pseudoinverse Jacobian method of redundancy resolution. The three subtasks are end effector trajectory represented by $\mathcal {T}_1$, hip trajectory represented by $\mathcal {T}_2$, and maximizing the manipulability represented by $\mathcal {T}_3$. The Pseudoinverse Jacobian method is coupled with particle swarm optimization (PSO) to find the optimal trajectories. The main contribution of this paper is the decomposition of the whole-body task of the humanoid robot into three distinct subtasks to find energy-optimal, dynamically balanced, and obstacle-free trajectories. The concept of virtual surface is used to avoid dragging objects on the table surface. The problem is optimized with Particle Swarm Optimization. Simulations were conducted to pick up and place objects from a table and constrained spaces like a drawer. The results show that the robot can pick and place objects from defined locations on the table.

With the rapid advancements in robotics and autonomous driving, SLAM (simultaneous localization and mapping) has become a crucial technology for real-time localization and map creation, seeing widespread application across various domains. However, SLAM’s performance in dynamic environments is often compromised due to the presence of moving objects, which can introduce errors and inconsistencies in localization and mapping. To overcome these challenges, this paper presents a visual SLAM system that employs dynamic feature point rejection. The system leverages a lightweight YOLOv7 model for detecting dynamic objects and performing semantic segmentation. Additionally, it incorporates optical flow tracking and multiview geometry techniques to identify and eliminate dynamic feature points. This approach effectively mitigates the impact of dynamic objects on the SLAM process, while maintaining the integrity of static feature points, ultimately enhancing the system’s robustness and accuracy in dynamic environments. Finally, we evaluate our method on the TUM RGB-D dataset and in real-world scenarios. The experimental results demonstrate that our approach significantly reduces both the root mean square error (RMSE) and standard deviation (Std) compared to the ORB-SLAM2 algorithm.

In recent years, passive motion paradigms (PMPs), derived from the equilibrium point hypothesis and impedance control, have been utilised as manipulation methods for humanoid robots and robotic manipulators. These paradigms are typically achieved by creating a kinematic chain that enables the manipulator to perform goal-directed actions without explicitly solving the inverse kinematics. This approach leverages a kinematic model constructed through the training of artificial neural networks, aligning well with principles of cybernetics and cognitive computation by enabling adaptive and flexible control. Specifically, these networks model the relationship between joint angles and end-effector positions, facilitating the computation of the Jacobian matrix. Although this method does not require an accurate robot model, traditional neural networks often suffer from drawbacks such as overfitting and inefficient training, which can compromise the accuracy of the final PMP model. In this paper, we implement the method using a deep neural network and investigate the impact of activation functions and network depth on the performance of the kinematic model. Additionally, we propose a transfer learning approach to fine-tune the pre-trained model, enabling it to be transferred to other manipulator arms with different kinematic properties. Finally, we implement and evaluate the deep neural network-based PMP on the Universal Robots, comparing it with traditional kinematic controllers and assessing its physical interaction capabilities and accuracy.

Due to the effects of tolerance, design, and manufacturing deviations, there are clearances in the revolute joints of mechanical arms. These clearances can easily lead to system impacts and vibrations, resulting in a decrease in dynamic performance and affecting the trajectory tracking accuracy of the end effector. The existing dynamic models of mechanisms with clearance in revolute joints lack comprehensiveness, universality, and systematicity, and have not addressed the impact of joint reaction forces within clearance revolute joints on the system. The impact collision problem of the revolute joints with clearance was systematically, accurately, and comprehensively modeled and simulated in this study based on multibody dynamics theory. Based on Hertz’s elastic theory, the LuGre friction model, and joint reaction forces, this paper constructs constraint and mechanical models of revolute joints with clearance based on the theory of multibody dynamics. To facilitate multibody dynamics analysis, the collision impact direction matrix is proposed and used for the first time to transform the mechanical model of revolute joints with clearance into external forces. The dynamic models of mobile parallel and double serial manipulators are then constructed. Through numerical simulations on different clearance amounts, tracking trajectories, and load parameters, the impact of revolute joint clearances on system dynamic performance is analyzed. The engineering significance of this research in dynamic analysis of mobile parallel manipulators under imperfect revolute joint conditions is also discussed.

The study presents a novel approach to address challenges posed by singularities in robotic arm motion, focusing on Cartesian path planning and geometric path adherence. Recognizing limitations in traditional singularity avoidance methods, the research proposes a comprehensive strategy: reconstructing motion patterns in singular regions through singularity-consistent representations, applying arc-length reparameterization to Cartesian geometric paths, and incorporating path curvature as a dynamic weighting factor for sampling interval adjustment. This method achieves a balance between joint velocity smoothness and geometric tracking accuracy in Cartesian space, significantly enhancing the robot’s ability to adhere to prescribed geometric paths, particularly near singularities. Experimental results demonstrate the efficacy of the proposed approach in facilitating smooth singularity transitions, improving joint velocity continuity, and enhancing geometric path adherence. The study contributes to robotic arm path planning by offering a practical solution for applications requiring precise trajectory following and effective singularity handling.

This study focuses on the development and testing of SensHB.Q, a force-sensitive interface for driving omnidirectional motorized systems such as wheelchairs, precision agriculture rovers, hospital beds, mobile service robots, and heavy-duty platforms. As manual driving is still fundamental in transportation vehicles, the design of intuitive driving interfaces has a major impact on the user experience. Force-sensitive interfaces measure the forces and torques applied by the driver on a sensitive area of the device and then convert these force inputs into commands for the omnidirectional system. In this paper, the design of the SensHB.Q force-sensitive interface and the transfer function for converting force inputs into speed commands are presented. Experimental tests are then conducted to validate the effectiveness of this interface in controlling two omnidirectional motorized systems: MoviWE.Q electrically powered wheelchair and Agrimaro.Q rover for precision agriculture in greenhouses.

The estimation of workspace for parallel kinematic machines (PKMs) typically relies on geometric considerations, which is suitable for PKMs operating under light load conditions. However, when subjected to heavy load, PKMs may experience significant deformation in certain postures, potentially compromising their stiffness. Additionally, heavy load conditions can impact motor loading performance, leading to inadequate motor loading in specific postures. Consequently, in addition to geometric constraints, the workspace of PKMs under heavy load is also constrained by mechanism deformation and motor loading performance.

This paper aims at developing a new heavy load 6-PSS PKM for multi-degree of freedom forming process. Additionally, it proposes a new method for estimating the workspace, which takes into account both mechanism deformation and motor loading performance. Initially, the geometric workspace of the machine is predicted based on its geometric configuration. Subsequently, the workspace is predicted while considering the effects of mechanism deformation and motor loading performance separately. Finally, the workspace is synthesized by simultaneously accounting for both mechanism deformation and motor loading performance, and a new index of workspace utilization rate is proposed. The results indicate that the synthesized workspace of the machine diminishes as the load magnitude and load arm increase. Specifically, under a heavy load magnitude of 6000 kN and a load arm of 200 mm, the utilization rate of the synthesized workspace is only 9.9% of the geometric workspace.

Flapping-wing robots, inspired by natural flyers, have gained significant attention for surveillance and environmental monitoring applications. This study presents the design and analysis of a bat-inspired flapping-wing robot with foldable wings, aiming to enhance flight efficiency and maneuverability. The robot features silicone-based, stretchable membrane wings, with a wingspan of 1.4 m and a total mass of 620 g. A one-degree-of-freedom (DOF) revolute-spherical-spherical-revolute mechanism is used to reproduce the flapping motion, while a one-DOF Watt six-bar linkage mechanism enables dynamic wing folding, allowing adaptive wing shape modulation during flight. Explicit solutions for joint angle of the wing were expressed through analytical method. Flight tests were conducted to validate the effectiveness of the flapping-folding mechanism. Results show that the robot successfully replicates bat wing kinematics, with folding during the upstroke and unfolding during the downstroke. This research offers insights into bio-inspired wing designs for next-generation flapping-wing robots.

A transverse ledge climbing robot inspired by athletic locomotion is a customized robot aiming to travel through horizontal ledges in vertical walls. Due to the safety issue and complex configurations in graspable ledges such as horizontal, inclined ledges, and gaps between ledges, existing well-known vision-based navigation methods suffering from occlusion problems may not be applicable to this special kind of application. This study develops a force feedback-based motion planning strategy for the robot to explore and make feasible grasping actions as it continuously travels through reachable ledges. A contact force detection algorithm developed using a momentum observer approach is implemented to estimate the contact force between the robot’s exploring hand and the ledge. Then, to minimize the detection errors due to dynamic model uncertainties and noises, a time-varying threshold is integrated. When the estimated contact force exceeds the threshold value, the robot control system feeds the estimated force into the admittance controller to revise the joint motion trajectories for a smooth transition. To handle the scenario of gaps between ledges, several ledge-searching algorithms are developed to allow the robot to grasp the next target ledge and safely overcome the gap transition. The effectiveness of the proposed motion planning and searching strategy has been justified by simulation, where the four-link transverse climbing robot successfully navigates through a set of obstacle scenarios modeled to approximate the actual environment. The performance of the developed grasping ledge searching methods for various obstacle characteristics has been evaluated.

The human hand’s exceptional dexterity and compliance, derived from its rigid-soft coupling structure and tendon-driven interphalangeal coordination, inspire robotic grippers capable of versatile grasping and force adaptation. Traditional rigid manipulators lack compliance for delicate tasks, while soft robots often suffer from instability and low load capacity. To bridge this gap, we propose a biomimetic multi-joint composite finger integrating a 3D-printed rigid phalanges (46–51 mm) with dual fabric-reinforced pneumatic bladders, mimicking human finger biomechanics. This hybrid design combines hinge-jointed rigidity and anisotropic fabric constraints, enabling two rotational degrees of freedom with higher radial stiffness, achieving 2.18× higher critical burst pressure (240 kPa) than non-reinforced bladders, while preserving axial compliance. Experimental validation demonstrates a 4.77 N maximum fingertip force at 200 kPa and rapid recovery (< 2s) post-impact. The composite finger exhibits human-like gestures (enveloping, pinching, flipping) and adapts to irregular/fragile objects (e.g., eggs, screws) through coordinated bladder actuation. Assembled into a modular gripper, it sustains 1 kg payloads and executes thin-object flipping via proximal-distal joint synergy. This rigid-soft coupling design bridges compliance and robustness, offering high environmental adaptability for applications in industrial automation, human–robot interaction, and delicate manipulation.

The mandible is crucial for human physiological functions, as well as facial esthetics and expressions. The mandibular reconstruction surgery has dual challenges of restoration of both facial form and physiological function, which demands high precision in positioning and orientation of the bone graft. The traditional manual surgery heavily relies on surgeon’s experience. Although the computer image-guided surgery improves the positioning accuracy, the manual manipulation is still difficult to achieve precise spatial orientation of objects, resulting in unsatisfactory intraoperative execution of preoperative surgical design. This paper integrates computer image navigation and robotic technology to assist mandible reconstruction surgery, which empowers surgeons to achieve precise spatial localization and orientation adjustment of bone grafts. The kinematic analysis is conducted, and an improved Iterative Closest Point (ICP) algorithm is proposed for spatial registration. A novel hand-eye calibration method for multi-arm robot and spatial registration of free bone blocks are proposed. The precision experiment of the image-guided navigation and the animal experiments are carried out. The impact of registration point numbers on spatial registration accuracy is analyzed. The results show the feasibility of the robot-assisted navigation for mandibular reconstruction surgery. The robotic system can improve the orientation accuracy of bone blocks to enhance the effectiveness of surgery.

Optical microrobots are activated by a laser in a liquid medium using optical tweezers. To create visual control loops for robotic automation, this work describes a deep learning-based method for orientation estimation of optical microrobots, focusing on detecting 3-D rotational movements and localizing microrobots and trapping points (TPs). We integrated and fine-tuned You Only Look Once (YOLOv7) and Deep Simple Online Real-time Tracking (DeepSORT) algorithms, improving microrobot and TP detection accuracy by $\sim 3$% and $\sim 11$%, respectively, at the 0.95 Intersection over Union (IoU) threshold in our test set. Additionally, it increased mean average precision (mAP) by 3% at the 0.5:0.95 IoU threshold during training. Our results showed a 99% success rate in trapping events with no false-positive detection. We introduced a model that employs EfficientNet as a feature extractor combined with custom convolutional neural networks (CNNs) and feature fusion layers. To demonstrate its generalization ability, we evaluated the model on an independent in-house dataset comprising 4,757 image frames, where microrobots executed simultaneous rotations across all three axes. Our method provided mean rotation angle errors of $1.871^\circ$, $2.308^\circ$, and $2.808^\circ$ for X (yaw), Y (roll), and Z (pitch) axes, respectively. Compared to pre-trained models, our model provided the lowest error in the Y and Z axes while offering competitive results for X-axis. Finally, we demonstrated the explainability and transparency of the model’s decision-making process. Our work contributes to the field of microrobotics by providing an efficient 3-axis orientation estimation pipeline, with a clear focus on automation.