A deep learning-enabled human-cyber-physical fusion method towards human-robot collaborative assembly

Human-robot collaborative (HRC) assembly has become popular in recent years. It takes full advantage of the strength, repeatability and accuracy of robots and the high-level cognition, flexibility and adaptability of humans to achieve an ergonomic working environment with better overall productivity. However, HRC assembly is still in its infancy nowadays. How to ensure the safety and efficiency of HRC assembly while reducing assembly failures caused by human errors is challenging. To address the current challenges, this paper proposes a novel human-cyber-physical assembly system (HCPaS) framework, which combines the powerful perception and control capacity of digital twin with the virtual-reality interaction capacity of augmented reality (AR) to achieve a safe and efficient HRC environment. Based on the framework, a deep learning-enabled fusion method of HCPaS is proposed from the perspective of robot-level fusion and part-level fusion. Robot-level fusion perceives the pose of robots with the combination of PointNet and iterative closest point (ICP) algorithm, where the status of robots together with their surroundings could be registered into AR environment to improve the human's cognitive ability of complex assembly environment, thus ensuring the safe HRC assembly. Part-level fusion recognizes the type and pose of parts being assembled with a parallel network that takes an extended Pixel-wise Voting Network (PVNet) as the base architecture, on which assembly sequence/process information of the part could be registered into AR environment to provide smart guidance for manual work to avoid human errors. Eventually, experimental results demonstrate the effectiveness and efficiency of the approach.     

基于局部流形注意力的人脸表情识别

人脸表情识别在各种人机交互场景中有广泛的应用, 但在表情模糊或存在遮挡情况下, 现有的表情识别方法效果并不理想. 针对表情模糊和遮挡问题, 本文提出了一种基于局部流形注意力(SPD-Attention)的网络架构, 利用流形学习得到具有更强描述能力的二阶统计信息以加强对表情细节特征的学习, 抑制遮挡区域无关特征对网络的影响. 同时, 针对流形学习过程中由于对数计算导致的梯度消失和爆炸, 本文提出了相应的正则约束加速网络收敛. 本文在公开表情识别数据集上测试了算法效果, 与VGG等经典方法相比取得了显著提升, 在AffectNet、CK+、FER2013、FER2013plus、RAF-DB、SFEW上正确率分别为: 57.10%、99.01%、69.51%、87.90%、86.63%、49.18%, 并在模糊、遮挡表情数据集上相比于Covariance Pooling等目前先进方法提升了1.85%.     

多约束条件下机器人柔性装配技能自学习

机器人的装配策略受装配对象特性、装配工艺和装配控制方法的约束,针对装配过程接触阶段的位姿不确定性问题,提出一种装配姿态调整技能自学习的方法.首先描述多约束条件下的机器人装配技能问题,建立基于力/力矩、位姿、关节角度等多模信息描述的装配系统模型;然后构建融合竞争架构的机器人决策网络和策略优化网络,通过与环境的不断交互,进...     

The integration of local government,residents, and insurance in coastal adaptation: An agent-based modeling approach

Previous evaluation of flood risk overlooks the behavior and capacity of private stakeholders, thus limiting the application of adaptation policies. This study presents an agent-based model, applied to Miami-Dade County, FL, as the case study, to explore the public and private interaction in coastal flood adaptation and mitigation. The decision making of individuals' adaptive behavior is simulated based on the prospected theory under households' risk perception, insurance policies, and the local flood mitigation. The NFIP and private insurance policy are simulated separately to reflect the flood insurance market. Our results show that households' risk mitigation behaviors are clustered in high-risk coastal areas, including Miami-Beach and the east coast of the County. The overall flood risk is still high in the southern part of the County. To better reflect the flooding risk and address the affordability issue, a voucher coupled house elevation program could improve the insurance take-up rates as well as reduce the overall flood risk in the area. Results also indicate that private insurance would slightly increase if the NFIP's insurance rates increase. Afterward, four adaptation scenarios in response to future sea level rises are examined by considering the voucher-based insurance program and the local adaptation actions. Compared with the high-risk reduction but low coverage mitigation plan, more extensive coverage of public adaptation would better improve the overall adaptation outcome of the County, which indicates the importance of public participation in local risk mitigation and urban governance.     

基于延展性的机器人面料缝制张力预测方法

针对机器人缝制过程中未知面料期望张力导致面料变形的问题,根据面料的延展性分类训练支持向量机(SVM)模型,通过线性支持向量机模型预测未知面料的延展性;其次,采用模糊逻辑控制系统确定面料特性与期望张力之间的非线性关系;最后,利用毛呢、绸缎、天鹅绒布、摇粒绒布4种面料对SVM模型进行测试,通过模糊逻辑得到面料期望张力查询表,根据期望张力对毛呢和竹节棉麻进行缝制实验。结果表明:在拉伸面料的过程中,线性支持向量机模型预测的延展性最终趋于面料实际的延展性,基于模糊逻辑根据面料延展性和种类可实现对任意面料期望张力的预测,预测张力可满足智能化缝制加工的需要。该研究为避免面料形变从而提高缝制质量提供了前提条件。     

A novel offline programming approach of robot welding for multi-pipe intersection structures based on NSGA-Ⅱ and measured 3D point-clouds

The multi-pipe intersection structure in form of co-main pipe is widely used in various industries. To improve its welding quality and efficiency, this paper is devoted to proposing an offline programming approach to the robot trajectory based on NSGA-Ⅱ and measured 3D point-clouds. First, considering the existence of deviation between the actual workpiece and its ideal model, this paper selects the actual 3D point-cloud of weld seam as the research object and extracts its feature points by combining the characteristic of the spatial curve, which can reduce the data density while preserving their geometric characteristics. Second, to ensure the continuity of motion parameters while taking the calculation and fitting accuracy into account, the cubic NURBS is applied to fit the actual weld position, and a fast-adaptive fitting nodes configuration scheme is designed according to the variation characteristics of the spatial curve and the fitting error restriction. Third, this paper introduces a trajectory adaptive discretization method based on the chord error constraint for robot program generation, and establishes an optimization model of the robot welding trajectory based on NSGA-Ⅱ frame, which gives the robot joint's motion trajectories with optimal welding quality and cable twisting fluctuation. Finally, the experiments are designed to verify the correctness of the aforementioned approach.     

基于ANN的新型MOFs性能预测

在MOFs研究领域,探寻新型MOFs仍然是非常困难的研究问题.将MOFs进行"材料基因编码"后,应用遗传算法(Genetic Algorithm,GA)可以快速探索新型MOFs,但其性能依赖于设定的个体适应度函数,且对新生成的MOFs个体的有效评估也影响了该方法的效果.机器学习方法可以对MOFs的构效关系进行评估与预测,人工神经网络(Artificial Neural Network,ANN)是众多机器学习方法中具有代表性的一个,可以发掘非线性的构效关系.本文提出将神经网络用于预测遗传算法生成的新型MOFs个体对CH4气体的吸附能力,从而帮助遗传算法搜索新型MOFs.实验结果表明,神经网络可以有效评估新型MOFs材料,证明了将神经网络与遗传算法相结合用于新型MOFs搜索和筛选的可行性.     

A two-step method for kinematic parameters calibration based on complete pose measurement—Verification on a heavy-duty robot

The poor pose accuracy of industrial robots restricts their further application in aviation manufacturing. Kinematic calibration based on position errors is a traditional method to improve robot accuracy. However, due to the difference between length errors and angle errors in the order of magnitude, it is difficult to accurately calibrate these geometric parameters together. In this paper, a two-step method for robot kinematic parameters calibration and a novel method for position and orientation measurement are proposed and combined to identify these two kinds of errors respectively. The redundant parameter errors that affect the identification are also analyzed and eliminated to further improve the accuracy of this two-step method. Taking the Levenberg-Marquardt algorithm as the underlying algorithm, simulation results indicate that the proposed two-step calibration method has faster iteration speed and higher identification accuracy than the traditional one. On this basis, the calibration and measurement methods proposed in this paper are verified on a heavy-duty robot used for fiber placement. Experimental results show that the mean absolute position error decreases from 0.9906 mm to 0.3703 mm after calibration by the proposed two-step calibration method with redundancy elimination. The absolute position accuracy has increased by 41.81% compared with the traditional method based on position errors only and 14.97% compared with the two-step calibration method without redundancy elimination. At the same time, the orientation errors after calibration are not more than 0.1485°, and the average of absolute errors is 0.0447.