حس لامسه، یادگیری مهارت و دستکاری چابک رباتیک

Tactile Sensing, Skill Learning and Robotic Dexterous Manipulation focuses on cross-disciplinary lines of research and groundbreaking research ideas in three research lines: tactile sensing, skill learning and dexterous control. The book introduces recent work about human dexterous skill representation and learning, along with discussions of tactile sensing and its applications on unknown objects’ property recognition and reconstruction. Sections also introduce the adaptive control schema and its learning by imitation and exploration. Other chapters describe the fundamental part of relevant research, paying attention to the connection among different fields and showing the state-of-the-art in related branches. The book summarizes the different approaches and discusses the pros and cons of each. Chapters not only describe the research but also include basic knowledge that can help readers understand the proposed work, making it an excellent resource for researchers and professionals who work in the robotics industry, haptics and in machine learning. Front Cover Tactile Sensing, Skill Learning, and Robotic Dexterous Manipulation Copyright Contents Contributors Preface Part I Tactile sensing and perception 1 GelTip tactile sensor for dexterous manipulation in clutter 1.1 Introduction 1.2 An overview of the tactile sensors 1.2.1 Marker-based optical tactile sensors 1.2.2 Image-based optical tactile sensors 1.3 The GelTip sensor 1.3.1 Overview 1.3.2 The sensor projective model 1.3.3 Fabrication process 1.4 Evaluation 1.4.1 Contact localization 1.4.2 Touch-guided grasping in a Blocks World environment 1.5 Conclusions and discussion Acknowledgment References 2 Robotic perception of object properties using tactile sensing 2.1 Introduction 2.2 Material properties recognition using tactile sensing 2.3 Object shape estimation using tactile sensing 2.4 Object pose estimation using tactile sensing 2.5 Grasping stability prediction using tactile sensing 2.6 Vision-guided tactile perception for crack reconstruction 2.6.1 Visual guidance for touch sensing 2.6.2 Guided tactile crack perception 2.6.3 Experimental setup 2.6.4 Experimental results 2.7 Conclusion and discussion References 3 Multimodal perception for dexterous manipulation 3.1 Introduction 3.2 Visual-tactile cross-modal generation 3.2.1 ``Touching to see'' and ``seeing to feel'' 3.2.2 Experimental results 3.3 Spatiotemporal attention model for tactile texture perception 3.3.1 Spatiotemporal attention model 3.3.2 Spatial attention 3.3.3 Temporal attention 3.3.4 Experimental results 3.3.5 Attention distribution visualization 3.4 Conclusion and discussion Acknowledgment References 4 Capacitive material detection with machine learning for robotic grasping applications 4.1 Introduction 4.1.1 Motivation 4.1.2 Concept 4.1.3 Related work 4.2 Basic knowledge 4.2.1 Capacitance perception 4.2.1.1 Sensing hardware 4.2.1.2 Signal processing 4.2.1.3 Capacitance spectroscopy 4.2.2 Classification for material detection 4.2.2.1 k-Nearest neighbors 4.2.2.2 Support vector machines 4.2.2.3 Random forest classifier 4.2.2.4 Feedforward neural networks 4.2.2.5 Convolutional neural networks 4.3 Methods 4.3.1 Data preparation 4.3.1.1 Raw data 4.3.1.2 Image generation 4.3.2 Classifier configurations 4.4 Experiments 4.5 Conclusion References Part II Skill representation and learning 5 Admittance control: learning from humans through collaborating with humans 5.1 Introduction 5.2 Learning from human based on admittance control 5.2.1 Learning a task using dynamic movement primitives 5.2.1.1 Constructing a second-order nonlinear system 5.2.1.2 Learning the DMPs model 5.2.2 Admittance control model 5.2.3 Learning of compliant movement profiles based on biomimetic control 5.2.3.1 Robotic compliant movement representation 5.2.3.2 Adaptation law 5.3 Experimental validation 5.3.1 Simulation task 5.3.2 Handover task 5.3.3 Sawing task 5.4 Human robot collaboration based on admittance control 5.4.1 Principle of human arm impedance model 5.4.2 Estimation of stiffness matrix 5.4.3 Stiffness mapping between human and robot arm 5.5 Variable admittance control model 5.6 Experiments 5.6.1 Test of variable admittance control 5.6.2 Human–robot collaborative sawing task 5.7 Conclusion References 6 Sensorimotor control for dexterous grasping – inspiration from human hand 6.1 Introduction of sensorimotor control for dexterous grasping 6.2 Sensorimotor control for grasping kinematics 6.3 Sensorimotor control for grasping kinetics 6.4 Conclusions Acknowledgments References 7 From human to robot grasping: force and kinematic synergies 7.1 Introduction 7.1.1 Human hand synergies 7.1.2 The impact of the synergies approach on robotic hands 7.2 Experimental studies 7.2.1 Study 1: force synergies comparison between human and robot hands 7.2.2 Results of force synergies study 7.2.3 Study 2: kinematic synergies in both human and robot hands 7.2.4 Results of kinematic synergies study 7.3 Discussion 7.3.1 Force synergies: human vs. robot 7.3.2 Kinematic synergies: human vs. robot 7.4 Conclusions Acknowledgments References 8 Learning form-closure grasping with attractive region in environment 8.1 Background 8.2 Related work 8.2.1 Closure properties 8.2.2 Environmental constraints 8.2.3 Learning to grasp 8.3 Learning a form-closure grasp with attractive region in environment 8.3.1 Attractive region in environment for four-pin grasping 8.3.2 Learning to evaluate grasp quality with ARIE 8.3.2.1 Formation of the grasp quality measurement function 8.3.2.2 Uncertainties during grasping process 8.3.2.3 Calculation of the grasp quality score 8.3.2.4 Training of the network for grasp quality measurement 8.3.3 Learning to grasp with ARIE 8.3.3.1 Formation of the learning pipeline 8.3.3.2 Learning policy 8.3.3.3 Reward 8.4 Conclusion References 9 Learning hierarchical control for robust in-hand manipulation 9.1 Introduction 9.2 Related work 9.3 Methodology 9.3.1 Hierarchical structure for in-hand manipulation 9.3.2 Low-level controller 9.3.3 Mid-level controller 9.4 Experiments 9.4.1 Training mid-level policies and baseline 9.4.2 Dataset 9.4.3 Reaching desired object poses 9.4.4 Robustness analysis 9.4.5 Manipulating a cube 9.5 Conclusion References 10 Learning industrial assembly by guided-DDPG 10.1 Introduction 10.2 From model-free RL to model-based RL 10.2.1 Guided policy search 10.2.2 Deep deterministic policy gradient 10.2.3 Comparison of DDPG and GPS 10.3 Guided deep deterministic policy gradient 10.4 Simulations and experiments 10.4.1 Parameter lists 10.4.2 Simulation results 10.4.2.1 Comparison of different supervision methods 10.4.2.2 Effects of the supervision weight wto 10.4.2.3 Comparison of different algorithms 10.4.2.4 Adaptability of the learned policy 10.4.3 Experimental results 10.5 Chapter summary References Part III Robotic hand adaptive control 11 Clinical evaluation of Hannes: measuring the usability of a novel polyarticulated prosthetic hand 11.1 Introduction 11.2 Preliminary study 11.2.1 Data collection 11.2.1.1 Questionnaire 11.2.1.2 Focus group 11.2.2 Outcomes 11.3 The Hannes system 11.3.1 Analysis of survey study and definition of requirements 11.3.2 System architecture 11.3.2.1 The Hannes hand 11.3.2.2 Custom EMG sensors 11.4 Pilot study for evaluating the Hannes hand 11.4.1 Materials and methods 11.4.1.1 Subjects 11.4.1.2 Study protocol 11.4.1.3 Clinical evaluation measures 11.4.2 Results 11.5 Validation of custom EMG sensors 11.5.1 Materials and methods 11.5.1.1 Subjects 11.5.1.2 Study protocol 11.5.1.3 Analysis 11.5.2 Results 11.6 Discussion and conclusions References 12 A hand-arm teleoperation system for robotic dexterous manipulation 12.1 Introduction 12.2 Problem formulation 12.3 Vision-based teleoperation for dexterous hand 12.3.1 Transteleop 12.3.2 Pair-wise robot–human hand dataset generation 12.4 Hand-arm teleoperation system 12.5 Transteleop evaluation 12.5.1 Network implementation details 12.5.2 Transteleop evaluation 12.5.3 Hand pose analysis 12.6 Manipulation experiments 12.7 Conclusion and discussion References 13 Neural network-enhanced optimal motion planning for robot manipulation under remote center of motion 13.1 Introduction 13.2 Problem statement 13.2.1 Kinematics modeling 13.2.2 RCM constraint 13.2.2.1 2D RCM constraint 13.2.2.2 3D RCM constraint 13.3 Control system design 13.3.1 Controller design method 13.3.2 RBFNN-based approximation 13.3.3 Control framework 13.4 Simulation results 13.5 Conclusion References 14 Towards dexterous in-hand manipulation of unknown objects 14.1 Introduction 14.2 State of the art 14.3 Reactive object manipulation framework 14.3.1 Local manipulation controller – position part 14.3.2 Local manipulation controller – force part 14.3.3 Local manipulation controller – composite part 14.3.4 Regrasp planner 14.4 Finding optimal regrasp points 14.4.1 Grasp stability and manipulability 14.4.2 Object surface exploration controller 14.5 Evaluation in physics-based simulation 14.5.1 Local object manipulation 14.5.2 Large-scale object manipulation 14.6 Evaluation in a real robot experiment 14.6.1 Unknown object surface exploration by one finger 14.6.2 Unknown object local manipulation by two fingers 14.7 Summary and outlook Acknowledgment References 15 Robust dexterous manipulation and finger gaiting under various uncertainties 15.1 Introduction 15.2 Dual-stage manipulation and gaiting framework 15.3 Modeling of uncertain manipulation dynamics 15.3.1 State-space dynamics 15.3.2 Combining feedback linearization with modeling 15.4 Robust manipulation controller design 15.4.1 Design scheme 15.4.2 Design of weighting functions 15.4.2.1 Design of performance weighting function Wperf 15.4.2.2 Design of action weighting function Wu 15.4.2.3 Design of disturbance weighting function Wdis 15.4.2.4 Design of noise weighting function Wn 15.4.3 Manipulation controller design 15.5 Real-time finger gaits planning 15.5.1 Grasp quality analysis 15.5.2 Position-level finger gaits planning 15.5.3 Velocity-level finger gaits planning 15.5.4 Similarities between position-level and velocity-level planners 15.5.5 Finger gaiting with jump control 15.6 Simulation and experiment studies 15.6.1 Simulation setup 15.6.2 Experimental setup 15.6.3 Parameter lists 15.6.3.1 RMC parameters for simulation test 15.6.3.2 RMC parameters for BarrettHand experiment 15.6.3.3 Parameters for finger gaits planner simulation 15.6.4 RMC simulation results 15.6.4.1 Comparison with different methods 15.6.4.2 Robustness to uncertainties 15.6.5 RMC experiment results 15.6.6 Finger gaiting simulation results 15.6.6.1 Finger gaiting on smooth surfaces under uncertainties 15.6.6.2 Finger gaiting of a three-fingered hand 15.7 Chapter summary References A Key components of dexterous manipulation: tactile sensing, skill learning, and adaptive control A.1 Introduction A.2 Why sensing, why tactile sensing A.3 Why skill learning A.4 Why adaptive control A.5 Conclusion Index Back Cover Tactile Sensing, Skill Learning And Robotic Dexterous Manipulation Focuses On The Cross-disciplinary Lines Of Research And Ground-breaking Research Ideas On Three Research Lines: Tactile Sensing, Skill Learning And Dexterous Control. The Book Introduces The Recent Work About Human's Dexterous Skill Representation And Learning; Tactile Sensing And Its Applications On Unknown Objects' Property Recognition And Reconstruction. It Also Introduces The Adaptive Control Schema And Its Learning By Imitation And Exploration. The Book Describes The Fundamental Part Of Relevant Research, Paying Attention To The Connection Among Different Fields And Showing The State-of-the-art In The Related Branches. The Book Summarizes The Different Approaches And Discusses The Pros And Cons Of Each, And The Chapters Not Only Describe The Research But Also Include Basic Knowledge Which Can Help Readers To Understand The Proposed Work. It Also Gives Insight Into Recent Representative Results From Different Research Branches, A Whole Picture About The State-of-the-art And Potential Future Research Directions For Robotic Dexterous Manipulation. It Reveals And Illustrates How Robots Can Improve Its Dexterity By Modern Tactile Sensing, Interactive Perception, Learning And Adaptive Control Approaches. This Book Is An Excellent Resource For Researchers And Professionals Who Work In The Robotics Industry, Haptics And Machine Learning. Provides A Review Of Tactile Perception And Latest Advances In The Use Of Robotic Dexterous Manipulation Provides The Most Detailed Work On Synthesizing Intelligent Tactile Perception, Skill Learning And Adaptive Control Introduces The Recent Work On Human's Dexterous Skill Representation And Learning, The Adaptive Control Schema And Its Learning By Imitation And Exploration And The Adaptive Control Schema And Its Learning By Imitation And Exploration Reveals And Illustrates How Robots Can Improve Its Dexterity By Modern Tactile Sensing, Interactive Perception, Learning And Adaptive Control Approaches Tactile Sensing, Skill Learning and Robotic Dexterous Manipulation focuses on cross-disciplinary lines of research and groundbreaking research ideas in three research lines: tactile sensing, skill learning and dexterous control. The book introduces recent work about human dexterous skill representation and learning, along with discussions of tactile sensing and its applications on unknown objects'property recognition and reconstruction. Sections also introduce the adaptive control schema and its learning by imitation and exploration. Other chapters describe the fundamental part of relevant research, paying attention to the connection among different fields and showing the state-of-the-art in related branches. The book summarizes the different approaches and discusses the pros and cons of each. Chapters not only describe the research but also include basic knowledge that can help readers understand the proposed work, making it an excellent resource for researchers and professionals who work in the robotics industry, haptics and in machine learning. Provides a review of tactile perception and the latest advances in the use of robotic dexterous manipulation Presents the most detailed work on synthesizing intelligent tactile perception, skill learning and adaptive control Introduces recent work on human's dexterous skill representation and learning and the adaptive control schema and its learning by imitation and exploration Reveals and illustrates how robots can improve dexterity by modern tactile sensing, interactive perception, learning and adaptive control approaches