International Journal of Intelligent Robotics and Applications最新文献

The application of Human–Robot Collaboration (HRC) in the manufacturing sector, especially in the material handling process, is aimed at improving productivity through robots actively working alongside humans. In this condition, the robots need to understand how to handle the objects by themselves according to user preferences with an autonomous system. However, there have been challenges in the aspect of teaching robots to autonomously identify object grasp positions only using an RGB camera due to the effect of camera perspective on object visualization for robots. Therefore, this study aimed to propose a simplified method on an RGB camera for autonomous object grasping in the material handling process and implement it for the HRC concept. The method used a prototype robot manipulator with a computer vision system for object detection. During the execution of object grasping, the robot achieved a success rate of 86% for a single object and 76% for multiple objects. In the HRC concept, the robot achieved a success rate of 92% for placing objects one by one and 84% for placing objects continuously. The result also showed fast inference time when the robot in real-time detected the object, which was even just running on the CPU and in the planning process without complexity and requiring additional equipment aside from an RGB camera.

The aging power plants are critical infrastructures that require regular inspection to ensure their longevity. Automated inspection, in particular, can save a significant amount of time and cost, and eliminate the safety concerns of manual inspection. For this purpose, we propose a biologically-inspired robot that integrates friction-based mobility and sensing. It is capable of traversing horizontal and vertical boiler tubes as well as tubes with a (45^{circ }) bend. Furthermore, the friction pads on the robot fingers allow for locomotion on the rough surfaces of the boiler tubes. These pads also provide grip on non-magnetic tubes enabling the robot to be deployed on tubes made of any material. In addition, this robot has electromagnetic acoustic transducers (EMAT) embedded in all of its fingers that enable defect detection during locomotion. The presented platform can inspect complex tubular structures and considerably reduce the time, cost, and hazards experienced in manual inspection.

Adaptability and robustness are the important expressions of intelligent walking ability of humanoid robots. However, they may be in an unstable state due to the huge impact contact forces produced by foot instant landing. This paper is concerned with the problem of dynamical biped walking and robust control of humanoid robots under ground reaction forces (GRF). In order to imitate human’s muscles to absorb the landing force, the robotic system is modeled as a mass–damp–spring model. The novelty of the article lies in the use of impedance control based on ground reaction forces, which deals with the complicated optimization problem subjected to both equality and inequality constraints. A feedback controller is designed to utilize inertial damping to generate the desired motion trajectory of the robot. The constructing autonomous evolution mechanism is mentioned to realize adaptive optimization of walking model. It ensures that the impact of GRF and reinforce stability during transition from single support phase to double support phase. Finally, the effectiveness of the proposed method is verified by simulations.

Existing lidar-based semantic segmentation algorithms and datasets focus on autonomous vehicles operating in urban environments. This has greatly improved the safety and reliability of these autonomous vehicles in predictable scenery. A new dataset provides lidar data focusing on off-road environments as seen by autonomous ground vehicles, ushering in a new era of off-road exploration capabilities. To the best of our knowledge, no new algorithms have been developed specifically for this unstructured environment. To gain an understanding of how existing algorithms perform in an off-road environment, we assess the baseline performance of four algorithms, KPConv, SalsaNext, Cylinder3D, and SphereFormer, on a commonly used on-road dataset, SemanticKITTI. We then compare the results with an off-road dataset, RELLIS-3D. We discuss the degradation of each algorithm on the off-road dataset and investigate potential causes such as class imbalance, inconsistencies in the labeled data, and the inherent difficulty of segmenting off-road environments. We present the strengths and weaknesses of each algorithm’s segmentation abilities and provide a comparison of the runtime of each algorithm for real-time capabilities. This is crucial for identifying what network architecture features are potentially the most beneficial for unstructured scenes. A robust, open-source software implementation via docker containers and bash scripts provides simple, repeatable execution of all algorithm training and evaluations. All code is publicly available at https://github.com/UA-Lidar-Segmentation-Research.

This paper presents a novel approach to control and actuate a Continuous Passive Motion (CPM) machine by integrating a deep learning-based control strategy using convolutional neural networks in a gaming context for providing post-surgical therapy and knee rehabilitation. Electromyography and inertial measurement unit sensors are interfaced with the patient's thigh muscles to record the patient's intent signals and classify them as three states: forward, backward, and rest. Comparison studies have been performed to prove the novelty of the proposed lightweight convolutional neural network architecture over other architectures and machine learning methodologies for real-time implementation. Additionally, gaming software has been interfaced, making the recovery process motivating to deal with the psychological aspects of rehabilitation. A low-cost, ecofriendly alpha prototyped CPM machine is prototyped for implementing the algorithms. Experiments are performed on three healthy subjects to establish the feasibility of this home rehabilitation device under professional guidance. Thus, this study aims to improve home-based knee rehabilitation effectiveness, offering complete recovery to the patients, delivering intensive and motivational rehabilitation.

The drive system of robots and robot-like systems (RLS) is often designed with a combination of an e-motor and a gearbox with a high transmission ratio to optimize performance. The various types of possible robot gearboxes can be selected based on their characteristics, which strongly influence the performance of the entire robotic system. Planetary gear drives have advantages due to their high efficiency and low design complexity. Disadvantageous is the low transmission ratio per stage and the resulting large design space required with the currently predominant involute gearing. Using special tooth profile shapes, such as the eccentric cycloid (EC) gearing, enables a high transmission ratio per stage to be achieved, thus reducing the design space required. In order to evaluate the design, a description of the geometry and characteristics of the EC gearing is necessary. The application-optimized design can be made accessible on an interdisciplinary basis using a suitable description language for the product development of the complete robotic system. The paper shows the design and analysis of a planetary gearbox with a high transmission ratio for applications in robotics. The planetary gear stage is designed with the EC gearing, which offers advantages compared to the involute gearing. The performance of the selected gearing is evaluated based on various characteristics. This allows advantages to be identified compared to the established types of transmission for robots and RLS. Overall, the paper presents a new robot gearbox with a comprehensive description and analysis directly accessible for simulation or production using additive manufacturing.

A learning-based strategy for the trajectory tracking of redundant mobile manipulators (MM) was presented in this study. A five-degrees-of-freedom (DOF) manipulator is mounted on the differential drive (DD) mobile robot. The advantage of using a redundant system is to avoid joint limits, obstacles, and singularities towards desired trajectory tracking. The proposed approach is based on the Kohonen Self-Organizing Map (KSOM) advanced with Weighted Least Norm (WLN) matrix algorithm. This approach is the recommended neural network for inverse kinematics solutions because of its stability, preserved topology, and capacity to optimize the joint space trajectory while producing a smooth minimal joint angle. A proposed method for redundancy resolution in MM has been simulated using MATLAB simulation code and the Gazebo real-time simulation physical environment. The simulation results are evaluated with the joint limit method of redundancy resolution and other existing controllers for verification purposes. The conventional method of redundancy resolution is local optimum and infeasible for the end-effector motion in the entire workspace. The KSOM uses different steps of error correction that improve the system’s performance as well as ensure the global asymptotical stability of the system. The Root Mean Square Error (RMSE) values for straight-line, circular, Lissajious, and irregular sinusoidal path motions of the proposed method using KSOM are given as 0.0095 m, 0.009945 m, 0.009897 m, and 0.009758 m, respectively. The simulation results of the proposed method confirm the effectiveness of the proposed approach.

To enhance robotic manipulator adaptation to human partners and minimize human energy consumption in human–robot collaboration, this paper introduces a neural admittance control framework, which integrates human motion intention estimation and force feedforward compensation. Maximum likelihood estimation is employed to derive human motion intention and stiffness within human–robot collaboration, which are seamlessly merged into admittance control. Force feedforward compensation is proposed to minimize interaction force and position tracking fluctuations based on estimated human intention and stiffness. RBF neural network control is used to refine position inner loop dynamics and to improve position tracking accuracy and response speed. Comprehensive comparative simulations validate the method’s effectiveness in diminishing human–robot interaction force, enhancing position response speed, and mitigating interaction force and position fluctuations. The experiment performed on the Franka Emika Panda robot platform, illustrates that the proposed method is effective and enhance human-robot collaboration.

In our industrial material defect detecting processes, the multi criteria is considered in two-level motion planning structure. Firstly, the feed speed of the end-effector should be programmed in optimal time for satisfying the requirement of high efficiency. Secondly, the planned joint velocities and accelaration are characterized by high-order derivatives to guarantee smooth motion, taking into account the kinematic constraints. Last but not least, energy consumption of the robot’s movement is a focus during designing trajectories. The Pareto optimal method is applied to solve the trajectory planning problem. The results of the experiments suggest that the Pareto approach can realize effective multi-objective optimization and deliver a group of Pareto solutions for decision makers. Based on the actual requirements, suitable Pareto-optimal trajectory can be achieved and the practical operation of the industrial robot is good.