Prof. Qixin Cao, Shanghai Jiao Tong University, China

Title:  The Componentized Composite Robot and Autonomous Grasping Technology
Abstract: Compound robot is a new type of robot which combines mobile robot with industrial robot.  In the industrial field, industrial robots are called mechanical arm or manipulator, which is mainly used to replace the grasping function of manipulator.  The mobile robot, or Mecanum Wheel AGV, replaces the walking function of human legs and feet, and composite robots use their hands and feet to combine the two functions.  Compared with the manipulator, the compound robot has high flexibility and the ability to deal with the inherent uncertainty of the system. However, the complex control system and the lack of general robot control software architecture make it difficult to popularize.  This report introduces a composite robot based on component technology.  The composite robot can configure functional components according to the task to adapt to the uncertain environment and complete the autonomous visual grasping and dexterous operation. In this experiment, the hardware of a compound robot is composed of UR10 cooperative manipulator, Mecanum wheel mobile robot and RGBD sensor. The robot is required to have the functions of environment perception, object grasping and recognition, autonomous motion planning and visual control.  Using component technology to integrate slam, visual recognition, deep learning, with / without model grabbing recognition, independent planning operation algorithm, greatly improve the performance and reduce the development time.  The results of JD- X Robot Challenge Contest  show that the autonomous composite robot has high reliability for the autonomous system of grasping tasks with and with models.

Prof. Li Guo, Hunan University, China

Title: Research on Acoustic Emission Intelligent Monitoring of Grinding Engineering Ceramics
Abstract: Firstly, acoustic emission (AE) signal analysis by use of short time Fourier transform is used to monitor the grinding temperature of engineering ceramics. The relationship between the acoustic emission signal of high speed grinding of engineering ceramics and grinding force, grinding temperature and grinding damage are studied. Secondly, High precision AE monitoring of diamond grinding wheel wear in engineering ceramics grinding were carried out. The BP neural network for monitoring the passivation degree of diamond grinding wheel in high speed cermet grinding based on wavelet analysis of AE signal is studied. The variance of wavelet decomposition coefficient of AE signal in alumina grinding is used as the input feature of support vector machine. The empirical mode decomposition (EMD) of grinding AE signal is used to extract the effective value, variance and energy coefficient of its intrinsic mode function as the input features of least squares support vector machine. A high-precision detection method of grinding wheel loading in high speed grinding based on active infrared detection technology and thermal image segmentation technology is studied. Thirdly, AE monitoring of grinding mechanism of engineering ceramics is carried out. The frequency spectrum and eigenvalues of AE signals of single diamond abrasive scratching alumina and zirconia ceramics at different cutting depths are studied. The autoregressive moving average model can better characterize the AE signal characteristics of single diamond abrasive scratching alumina and zirconia ceramics, and can analyze the depth of single diamond abrasive scratching ceramics in real time. A method of monitoring the grinding contact between diamond wheel and zirconia ceramic specimen by using the energy of AE signal reconstructed by wavelet transform is proposed. High speed grinding force and AE signal of cemented carbide were studied. Fourthly, the optimized BP neural network is used to monitor the grinding surface roughness with high precision by use of AE. The research solved the problem of acoustic emission monitoring in engineering ceramics grinding and laid the foundation for its practical application!

Prof. Huailin Zhao, Shanghai Institute of Technology, China

Title: Deep Learning Based High Accuracy Crowd Counting
Abstract: Crowd counting is a computer vision problem, which estimates the total number of people in images or surveillance videos. The accurate crowd counting method is of great significance in crowd control, crowd analysis, video surveillance and public safety management. This presentation simply introduces the concept of crowd counting and its main methods based on the convolutional neural network(CNN). Then it stresses on how to improve the counting accuracy of the crowd image with high density or dramatic density changes. And it gives a few examples of CNN models for accurate crowd counting. At last, an Intuitive video demo of crowd counting is shown.