Muutke küpsiste eelistusi

Robot Programming by Demonstration [Kõva köide]

4.50/5 (4 hinnangut Goodreads-ist)
(Learning Algorithms and Systems Laboratory, Switzerland)
  • Formaat: Hardback, 320 pages, kaal: 612 g
  • Ilmumisaeg: 15-Apr-2021
  • Kirjastus: Taylor & Francis Inc
  • ISBN-10: 1439808678
  • ISBN-13: 9781439808672
Teised raamatud teemal:
Robot Programming by DemonstrationSuurem pilt
  • Formaat: Hardback, 320 pages, kaal: 612 g
  • Ilmumisaeg: 15-Apr-2021
  • Kirjastus: Taylor & Francis Inc
  • ISBN-10: 1439808678
  • ISBN-13: 9781439808672
Teised raamatud teemal:
Püsilink: https://www.kriso.ee/db/9781439808672.html
Märksõnad:

Also referred to as learning by imitation, tutelage, or apprenticeship learning, Programming by Demonstration (PbD) develops methods by which new skills can be transmitted to a robot. This book examines methods by which robots learn new skills through human guidance. Taking a practical perspective, it covers a broad range of applications, including service robots. The text addresses the challenges involved in investigating methods by which PbD is used to provide robots with a generic and adaptive model of control. Drawing on findings from robot control, human-robot interaction, applied machine learning, artificial intelligence, and developmental and cognitive psychology, the book contains a large set of didactic and illustrative examples. Practical and comprehensive machine learning source codes are available on the book’s companion website: http://www.programming-by-demonstration.org

Acknowledgment v
Introduction
1(30)
Contributions
3(1)
Organization of the book
3(1)
Review of Robot Programming by Demonstration (PBD)
4(8)
The birth of programmable machines
4(1)
Early work of PbD in software development
5(1)
Early work of PbD in robotics
6(1)
Toward the use of machine learning techniques in PbD
7(2)
From a simple copy to the generalization of a skill
9(1)
From industrial robots to service robots and humanoids
10(1)
From a purely engineering perspective to an interdisciplinary approach
11(1)
Current state of the art in PbD
12(19)
Human-robot interfaces
12(2)
Learning skills
14(8)
Incremental teaching methods
22(1)
Human-robot interaction in PbD
23(3)
Biologically-oriented learning approaches
26(5)
System Architecture
31(44)
Illustration of the proposed probabilistic approach
31(3)
Encoding of motion in a Gaussian Mixture Model (GMM)
34(1)
Recognition, classification and evaluation of a reproduction attempt
35(1)
Encoding of motion in Hidden Markov Model (HMM)
35(3)
Recognition, classification and evaluation of a reproduction attempt
37(1)
Reproduction through Gaussian Mixture Regression (GMR)
38(6)
Reproduction by considering multiple constraints
44(3)
Direct computation method
44(1)
Method based on optimization of a metric of imitation
45(2)
Learning of model parameters
47(8)
Batch learning of the GMM parameters
47(2)
Batch learning of the HMM parameters
49(2)
Incremental learning of the GMM parameters
51(4)
Reduction of dimensionality and latent space projection
55(5)
Principal Component Analysis (PCA)
56(1)
Canonical Correlation Analysis (CCA)
57(1)
Independent Component Analysis (ICA)
57(1)
Discussion on the different projection techniques
58(2)
Model selection and initialization
60(4)
Estimating the number of Gaussians based on the Bayesian Information Criterion (BIC)
60(1)
Estimating the number of Gaussians based on trajectory curvature segmentation
61(3)
Regularization of GMM parameters
64(5)
Bounding covariance matrices during estimation of GMM
64(1)
Single mode restriction during reproduction through GMR
64(3)
Temporal alignment of trajectories through Dynamic Time Warping (DTW)
67(2)
Use of prior information to speed up the learning process
69(2)
Extension to mixture models of varying density distributions
71(2)
Generalization of binary signals through a Bernoulli Mixture Model (BMM)
71(2)
Summary of the chapter
73(2)
Comparison and Optimization of the Parameters
75(26)
Optimal reproduction of trajectories through HMM and GMM/GMR
75(12)
Experimental setup
75(4)
Experimental results
79(8)
Optimal latent space of motion
87(5)
Experimental setup
87(2)
Experimental results
89(3)
Optimal selection of the number of Gaussians
92(2)
Experimental setup
93(1)
Experimental results
93(1)
Robustness evaluation of the incremental learning process
94(7)
Experimental setup
95(2)
Experimental results
97(4)
Handling of Constraints in Joint Space and Task Space
101(28)
Inverse kinematics
101(5)
Local solutions
102(2)
Extending inverse kinematics solutions to a statistical framework
104(2)
Handling of task constraints in joint space-experiment with industrial robot
106(10)
Experimental setup
109(4)
Experimental results
113(3)
Handling of task constraints in latent space-experiment with humanoid robot
116(13)
Experimental setup
120(1)
Experimental results
120(9)
Extension to Dynamical System and Handling of Perturbations
129(18)
Proposed dynamical system
130(5)
Extension to motions containing pauses and loops
133(2)
Influence of the dynamical system parameters
135(1)
Experimental setup
135(6)
Illustration of the problem
138(1)
Handling of multiple landmarks
139(1)
Handling of inverse kinematics
140(1)
Experimental results
141(6)
Transferring Skills Through Active Teaching Methods
147(24)
Experimental setup
148(3)
Experimental results
151(15)
Learning bimanual gestures
151(1)
Learning to stack objects
152(6)
Learning to move chess pieces
158(8)
Roles of an active teaching scenario
166(5)
Insights from psychology
166(1)
Insights from developmental sciences
167(2)
Insights from sociology
169(1)
Insights from sports science
169(2)
Using Social Cues to Speed up the Learning Process
171(10)
Experimental setup
173(5)
Use of head/gaze information as priors
173(3)
Use of vocal information as priors
176(2)
Experimental results
178(3)
Discussion, Future Work and Conclusions
181(20)
Advantages of the proposed approach
181(7)
Advantages of using motion sensors to track gestures
181(2)
Advantages of the HMM representation for imitation learning
183(2)
Advantages of the GMR representation for regression
185(3)
Failures and limitations of the proposed approach
188(6)
Loss of important information through PCA
188(1)
Failures at learning incrementally the GMM parameters
189(4)
Failures at extending a skill to a context that is too dissimilar to the ones encountered
193(1)
Further issues
194(4)
Toward combining exploration and imitation
194(1)
Toward a joint use of discrete and continuous constraints
195(3)
Toward predicting the outcome of a demonstration
198(1)
Final words
198(3)
References 201(20)
Index 221
Learning Algorithms and Systems Laboratory, Switzerland