一种用于机器人视觉系统的码链曲线角点寻求算法

本文描述了一种在平面码链曲线上检测尖锐角点的方法。对如何测量角点凸性进行了详细介绍。通过例子说明了该算法的实效性。最后,将计算法应用于机器人视觉系统,给出了仿真结果。     

Incremental topological flipping works for regular triangulations

A set ofn weighted points in general position in ^d defines a unique regular triangulation. This paper proves that if the points are added one by one, then flipping in a topological order will succeed in constructing this triangulation. If, in addition, the points are added in a random sequence and the history of the flips is used for locating the next point, then the algorithm takes expected time at mostO(nlogn+n ^[d/2]). Under the assumption that the points and weights are independently and identically distributed, the expected running time is between proportional to and a factor logn more than the expected size of the regular triangulation. The expectation is over choosing the points and over independent coin-flips performed by the algorithm.The research of both authors was supported by the National Science Foundation under Grant CCR-8921421 and the research by the first author was also supported under the Alan T. Waterman award, Grant CCR-9118874. Any opinions, findings, conclusions, or recommendations expressed in this publication are those of the authors and do not necessarily reflect the view of the National Science Foundation.     

Intelligent assistance for older adults via an admittance-controlled wheeled mobile manipulator with task-dependent end-effectors

The increase in the ageing population worldwide poses a severe challenge in assisting older individuals to live independently, including the provision of mobility assistance and support in daily activities. In this paper, a practical robotic system is developed to provide intelligent support for older persons using a wheeled mobile manipulator (WMM), consisting of an omnidirectional mobile platform and a robotic arm. We focus on two critical needs: (1) mobility assistance, and (2) object manipulation support. The tasks are not executed simultaneously and each uses a task-dependent end-effector. Learning from demonstration, or kinesthetic teaching, is adopted to help the WMM to learn an elderly or disabled user’s walking pattern or an able-bodied person’s object manipulation skill. The robotic system can assist the user in conducting a number of daily operations. For mobility assistance, the WMM is reconfigured into a smart walker, where a novel variable admittance control is adopted to detect the user’s walking intention. A learning approach based on dynamic movement primitives is implemented to capture and adapt the WMM to the user’s walking pattern. For object manipulation support, a demonstrator first collaborates with an elderly user to conduct the task, and then the WMM takes the role of the demonstrator to assist the user. The Gaussian mixture model and Gaussian mixture regression are used to learn and reproduce the demonstrator’s experience, respectively. The advantages and effectiveness of the proposed approach are experimentally demonstrated with a four-wheel omnidirectional WMM.     

深度学习在微机原理实践课程中的应用

“深度学习”强调学习者主动、批判地学习知识,通过深度加工来构建知识体系、实现迁移应用。在高校课堂上则体现在教学过程中教师的引导和跟踪、学习过程中学生的主动及创造。本研究在深入分析“深度学习”的基础上,在《微机原理及接口技术实践》课程中进行深度学习尝试,同时引入MOOC(大型开放式网络课程)教学模式,从教学内容、教学方法、考核措施等方面着手设计,增强学生实践能力、创新意识、创新精神及团队合作意识等综合能力,从而促进学生高阶认知能力、科创能力的发展,完成教育个性化需求的落实。     

Learning multiple instance deep representation for objects tracking

Object tracking has been widely used in various intelligent systems, such as pedestrian tracking, autonomous vehicles. To solve the problem that appearance changes and occlusion may lead to poor tracking performance, we propose a multiple instance learning (MIL) based method for object tracking. To achieve this task, we first manually label the first several frames of video stream in image level, which can indicate that whether a target object in the video stream. Then, we leverage a pre-trained convolutional neural network that has rich prior information to extract deep representation of target object. Since the location of the same object in adjacent frames is similar, we introduce a particle filter to predict the location of target object within a specific region. Comprehensive experiments have shown the effectiveness of our proposed method.     

Comparison of encrypted control approaches and tutorial on dynamic systems using Learning With Errors-based homomorphic encryption

Encrypted control has been introduced to protect controller data by encryption at the stage of computation and communication, by performing the computation directly on encrypted data. In this article, we first review and categorize recent relevant studies on encrypted control. Approaches based on homomorphic encryption, multi-party computation, and secret sharing are introduced, compared, and then discussed with respect to computational complexity, communication load, enabled operations, security, and research directions. We proceed to discuss a current challenge in the application of homomorphic encryption to dynamic systems, where arithmetic operations other than integer addition and multiplication are limited. We also introduce a Learning With Errors based homomorphic cryptosystem called “Gentry-Sahai-Waters” scheme and discuss its benefits that allow for recursive multiplication of encrypted dynamic systems, without use of computationally expensive bootstrapping techniques.     

ML-Agents驱动的足球竞技游戏设计与实现

在人工智能飞速发展的大背景下,游戏AI的革新亦走在技术的前沿并引领游戏产业发展。为了研究游戏AI产生更为高效的游戏策略,对游戏研发中创新上限的瓶颈进行突破,本文设计一款ML-Agents驱动的足球竞技游戏,通过游戏AI的学习过程增强游戏NPC的可适应性和拟人化。借助CINEMA 4D、Maya设计场景建模,使用Unity引擎构建游戏场景,运用Anaconda搭建虚拟训练环境,采用TensorFlow训练NPC的行为,应用ML-Agents插件构建程序化人格,训练并引导Agent产生所期望的行为。实践结果表明,采用该方案研发的足球竞技游戏案例产品,游戏AI产生了更具创新性、更具效能的行为模式。     

智慧农业病虫害巡检系统设计与实现

为了应对农作物生长过程中病虫害的影响,本文开发了一个病虫害巡检系统。该系统以STM32F407ZGT6、ARM Cortex-A为采集与控制单元核心,配以自制智能小车,实现农作物环境监测及视频图像采集。针对农作物图像,利用深度学习技术,对其病虫害情况进行快速无损识别。使用MQTT协议实现本地数据上传,并对云端数据转发和存储。设计网页和手机端,方便查看农作物生长环境状态、病虫害识别情况等。该系统运用到农业生产中,能够有效防止病虫害事件扩大,具有准确性高、检测速度快、稳定性强等优点。     

基于优化变分模态分解和核极限学习机的集装箱吞吐量预测

针对港口集装箱吞吐量数据的复杂性特征,提出基于优化变分模态分解（OVMD）和核极限学习机（KELM）的集装箱吞吐量短期混合预测模型。首先,用汉佩尔辨识法（HI）剔除原始时间序列中的异常值,并把预处理之后的序列通过OVMD分解为多个特征明显的子模态。然后,为提高预测效率,将分解后的子模态按照样本熵（SE）值的大小分成高频低幅、中频中幅和低频高幅三类;同时,借助KELM中携带的小波、高斯和线性核函数捕捉具有不同特征子模态的趋势。最后,把所有子模态的预测结果线性相加得到最终的预测结果。以深圳港的月度集装箱吞吐量数据为样本进行实验,所提模型的平均绝对误差（MAE）达到0.914?9,平均绝对百分比误差（MAPE）达到0.199%,均方根误差（RMSE）达到7.886?0,决定系数（R²）为0.994?4。与四种对比模型相比,所提出的模型在预测精度和效率上都具有一定的优势,同时克服了传统互补集成经验模态分解（CEEMD）和集成经验模态分解（EEMD）中容易出现的模态混叠问题以及极限学习机（ELM）中存在过拟合等问题,具有一定的实际应用潜力。     

基于区块链的联邦学习研究进展

联邦学习（FL）是一种能够实现用户数据不出本地的新型隐私保护学习范式。随着相关研究工作的不断深入,FL的单点故障及可信性缺乏等不足之处逐渐受到重视。近年来,起源于比特币的区块链技术取得迅速发展,它开创性地构建了去中心化的信任,为FL的发展提供了一种新的可能。对现有基于区块链的FL框架进行对比分析,深入讨论区块链与FL相结合所解决的FL重要问题,并阐述了基于区块链的FL技术在物联网（IoT）、工业物联网（IIoT）、车联网（IoV）、医疗服务等多个领域的应用前景。