Learning Control : Applications in Robotics and Complex Dynamical Systems

Description

Front Cover Learning Control Copyright Contents List of contributors 1 A high-level design process for neural-network controls through a framework of human personalities 1.1 Introduction 1.2 Background 1.2.1 The CMAC associative-memory neural network 1.2.2 Unbiased nonlinearities 1.2.3 Direct adaptive control in the presence of bias 1.2.4 A graphical model of personalities 1.2.5 A computer model of personalities 1.3 Proposed methods 1.3.1 Proposed learning law 1.3.2 Cost functional for optimization 1.3.3 Stability analysis 1.4 Results 1.4.1 Developing a design procedure 1.4.2 Two-link robotic manipulator 1.5 Conclusions 1.A References 2 Cognitive load estimation for adaptive human–machine system automation 2.1 Introduction 2.1.1 Human–machine automation 2.1.2 Cognitive load measures 2.1.3 Some applications 2.2 Noninvasive metrics of cognitive load 2.2.1 Pupil diameter 2.2.2 Eye-gaze patterns 2.2.3 Eye-blink patterns 2.2.4 Heart rate 2.3 Details of open-loop experiments 2.3.1 Unmanned vehicle operators 2.3.2 Memory recall tasks 2.3.3 Delayed memory recall tasks 2.3.4 Simulated driving 2.4 Conclusions and discussions 2.5 List of abbreviations References 3 Comprehensive error analysis beyond system innovations in Kalman filtering 3.1 Introduction 3.2 Standard formulation of Kalman filter after minimum variance principle 3.3 Alternate formulations of Kalman filter after least squares principle 3.4 Redundancy contribution in Kalman filtering 3.5 Variance of unit weight and variance component estimation 3.5.1 Variance of unit weight and posteriori variance matrix of (k) 3.5.2 Estimation of variance components 3.6 Test statistics 3.7 Real data analysis with multi-sensor integrated kinematic positioning and navigation 3.7.1 Overview 3.7.2 Results 3.8 Remarks References 4 Nonlinear control 4.1 System modeling 4.1.1 Linear systems 4.1.2 Nonlinear systems 4.2 Nonlinear control 4.2.1 Feedback linearization 4.2.2 Stability and Lyapunov stability 4.2.3 Sliding mode control 4.2.4 Backstepping control 4.2.5 Adaptive control 4.3 Summary References 5 Deep learning approaches in face analysis 5.1 Introduction 5.2 Face detection 5.2.1 Sliding window 5.2.2 Region proposal 5.2.3 Single shot 5.3 Pre-processing steps 5.3.1 Face alignment 5.3.1.1 Discriminative model fitting 5.3.1.2 Cascaded regression 5.3.2 Pose estimation 5.3.3 Face frontalization 5.3.3.1 2D/3D local texture warping 5.3.3.2 Generative adversarial networks (GAN) based 5.3.4 Face super resolution 5.4 Facial attribute estimation 5.4.1 Localizing the ROI 5.4.2 Modeling the relationships 5.5 Facial expression recognition 5.6 Face recognition 5.7 Discussion and conclusion Pose Illumination Occlusion Lack of data Overfitting Expressions Subjectivity Aging Low quality camera shooting References 6 Finite multi-dimensional generalized Gamma Mixture Model Learning for feature selection 6.1 Introduction 6.2 The proposed model 6.3 Parameter estimation 6.4 Model selection using the minimum message length criterion 6.4.1 Fisher information for a generalized Gamma mixture model 6.4.2 Prior distribution h() 6.4.3 Algorithm 6.5 Experimental results 6.5.1 Texture images 6.5.2 Shape images 6.5.3 Scene images 6.6 Conclusion References 7 Variational learning of finite shifted scaled Dirichlet mixture models 7.1 Introduction 7.2 Model specification 7.2.1 Shifted-scaled Dirichlet distribution 7.2.2 Finite shifted-scaled Dirichlet mixture model 7.3 Variational Bayesian learning 7.3.1 Parameter estimation 7.3.2 Determination of the number of components 7.4 Experimental result 7.4.1 Malaria detection 7.4.2 Breast cancer diagnosis 7.4.3 Cardiovascular diseases (CVDs) detection 7.4.4 Spam detection 7.5 Conclusion 7.A 7.B References 8 From traditional to deep learning: Fault diagnosis for autonomous vehicles 8.1 Introduction 8.2 Traditional fault diagnosis 8.2.1 Model-based fault diagnosis 8.2.2 Signal-based fault diagnosis 8.2.3 Knowledge-based fault diagnosis 8.3 Deep learning for fault diagnosis 8.3.1 Convolutional neural network (CNN) 8.3.2 Deep autoencoder (DAE) 8.3.3 Deep belief network (DBN) 8.4 An example using deep learning for fault detection 8.4.1 System dynamics and fault models 8.4.1.1 System dynamics 8.4.1.2 Fault models 8.4.2 Deep learning methodology 8.4.3 Fault classification results 8.5 Conclusion References 9 Controlling satellites with reaction wheels 9.1 Introduction 9.2 Spacecraft attitude mathematical model 9.2.1 Coordinate frame 9.2.2 Spacecraft dynamics 9.2.3 Attitude kinematics 9.2.4 External disturbances 9.3 Attitude tracking 9.4 Actuator dynamics 9.4.1 Simple brushless direct current motor 9.4.2 Mapping matrix 9.4.3 Reaction wheel parameters 9.5 Attitude control law 9.5.1 Basics of variable structure control 9.5.2 Design of sliding manifold 9.5.3 Control law 9.5.4 Stability analysis 9.6 Performance analysis 9.7 Conclusions References 10 Vision dynamics-based learning control 10.1 Introduction 10.2 Problem definition 10.2.1 Learning a vision dynamics model 10.3 Experiments 10.4 Conclusions References Index Back Cover