Download the full version and explore a variety of ebooks

or textbooks at https://ebookmass.com

Learning Control Zhang

_____ Follow the link below to get your download now _____

https://ebookmass.com/product/learning-control-zhang/

Access ebookmass.com now to download high-quality

ebooks or textbooks
Here are some recommended products for you. Click the link to
download, or explore more at ebookmass.com

Learning Control Dan Zhang

https://ebookmass.com/product/learning-control-dan-zhang/

MOTOR LEARNING AND CONTROL Concepts and Applications

Richard A. Magill

https://ebookmass.com/product/motor-learning-and-control-concepts-and-
applications-richard-a-magill/

Craving Control Faust

https://ebookmass.com/product/craving-control-faust/

Plasmonic MEMS John X. J. Zhang

https://ebookmass.com/product/plasmonic-mems-john-x-j-zhang/
Transcritical CO2 Heat Pump Xin-Rong Zhang

https://ebookmass.com/product/transcritical-co2-heat-pump-xin-rong-
zhang/

Intelligent Satellite Design and Implementation Jianjun

Zhang

https://ebookmass.com/product/intelligent-satellite-design-and-
implementation-jianjun-zhang/

Four Treasures of the Sky Jenny Tinghui Zhang

https://ebookmass.com/product/four-treasures-of-the-sky-jenny-tinghui-
zhang/

Physical Fundamentals of Nanomaterials 1st Edition Bangwei

Zhang

https://ebookmass.com/product/physical-fundamentals-of-
nanomaterials-1st-edition-bangwei-zhang/

Big Data Mining for Climate Change Zhihua Zhang

https://ebookmass.com/product/big-data-mining-for-climate-change-
zhihua-zhang/
Learning Control

Applications in Robotics and

Complex Dynamical Systems

FIRST EDITION

Dan Zhang
York University, Toronto, ON, Canada

Bin Wei
Algoma University, Sault Ste Marie, ON, Canada
Table of Contents

Cover image

Title page

Copyright

List of contributors

Chapter 1: A high-level design process for neural-network controls

through a framework of human personalities

Abstract

1.1. Introduction

1.2. Background

1.3. Proposed methods

1.4. Results

1.5. Conclusions

Appendix 1.A.

References
Chapter 2: Cognitive load estimation for adaptive human–machine
system automation

Abstract

2.1. Introduction

2.2. Noninvasive metrics of cognitive load

2.3. Details of open-loop experiments

2.4. Conclusions and discussions

2.5. List of abbreviations

References

Chapter 3: Comprehensive error analysis beyond system innovations

in Kalman filtering

Abstract

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

3.6. Test statistics

 3.7. Real data analysis with multi-sensor integrated kinematic
positioning and navigation

3.8. Remarks

References

Chapter 4: Nonlinear control

Abstract

4.1. System modeling

4.2. Nonlinear control

4.3. Summary

References

Chapter 5: Deep learning approaches in face analysis

Abstract

5.1. Introduction

5.2. Face detection

5.3. Pre-processing steps

5.4. Facial attribute estimation

5.5. Facial expression recognition

5.6. Face recognition

5.7. Discussion and conclusion

 References

Chapter 6: Finite multi-dimensional generalized Gamma Mixture

Model Learning for feature selection

Abstract

6.1. Introduction

6.2. The proposed model

6.3. Parameter estimation

6.4. Model selection using the minimum message length

criterion

6.5. Experimental results

6.6. Conclusion

References

Chapter 7: Variational learning of finite shifted scaled Dirichlet

mixture models

Abstract

7.1. Introduction

7.2. Model specification

7.3. Variational Bayesian learning

7.4. Experimental result

7.5. Conclusion
 Appendix 7.A.

Appendix 7.B.

References

Chapter 8: From traditional to deep learning: Fault diagnosis for

autonomous vehicles

Abstract

8.1. Introduction

8.2. Traditional fault diagnosis

8.3. Deep learning for fault diagnosis

8.4. An example using deep learning for fault detection

8.5. Conclusion

References

Chapter 9: Controlling satellites with reaction wheels

Abstract

9.1. Introduction

9.2. Spacecraft attitude mathematical model

9.3. Attitude tracking

9.4. Actuator dynamics

9.5. Attitude control law

 9.6. Performance analysis

9.7. Conclusions

References

Chapter 10: Vision dynamics-based learning control

Abstract

10.1. Introduction

10.2. Problem definition

10.3. Experiments

10.4. Conclusions

References

Index
Copyright
Elsevier
Radarweg 29, PO Box 211, 1000 AE Amsterdam, Netherlands
The Boulevard, Langford Lane, Kidlington, Oxford OX5 1GB, United
Kingdom
50 Hampshire Street, 5th Floor, Cambridge, MA 02139, United States

Copyright © 2021 Elsevier Inc. All rights reserved.

MATLAB® is a trademark of The MathWorks, Inc. and is used with

permission.
The MathWorks does not warrant the accuracy of the text or
exercises in this book.
This book's use or discussion of MATLAB® software or related
products does not constitute endorsement or sponsorship by The
MathWorks of a particular pedagogical approach or particular use of
the MATLAB® software.

No part of this publication may be reproduced or transmitted in any

form or by any means, electronic or mechanical, including
photocopying, recording, or any information storage and retrieval
system, without permission in writing from the publisher. Details on
how to seek permission, further information about the Publisher's
permissions policies and our arrangements with organizations such
as the Copyright Clearance Center and the Copyright Licensing
Agency, can be found at our website:
www.elsevier.com/permissions.
This book and the individual contributions contained in it are
protected under copyright by the Publisher (other than as may be
noted herein).

Notices
Knowledge and best practice in this field are constantly changing.
As new research and experience broaden our understanding,
changes in research methods, professional practices, or medical
treatment may become necessary.

Practitioners and researchers must always rely on their own

experience and knowledge in evaluating and using any
information, methods, compounds, or experiments described
herein. In using such information or methods they should be
mindful of their own safety and the safety of others, including
parties for whom they have a professional responsibility.

To the fullest extent of the law, neither the Publisher nor the
authors, contributors, or editors, assume any liability for any injury
and/or damage to persons or property as a matter of products
liability, negligence or otherwise, or from any use or operation of
any methods, products, instructions, or ideas contained in the
material herein.

Library of Congress Cataloging-in-Publication Data

A catalog record for this book is available from the Library of
Congress

British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library

ISBN: 978-0-12-822314-7

For information on all Elsevier publications visit our website at

https://www.elsevier.com/books-and-journals
Publisher: Matthew Deans
Acquisitions Editor: Dennis McGonagle
Editorial Project Manager: John Leonard
Production Project Manager: Poulouse Joseph
Designer: Victoria Pearson

Typeset by VTeX
List of contributors
Simge Akay Computer Engineering Department, Bahcesehir
University, Istanbul, Turkey
Basim Alghabashi Concordia Institute for Information Systems
Engineering (CIISE), Concordia University, Montreal, QC, Canada
Mohamed Al Mashrgy Electrical and Computer Engineering
(ECE), Al-Mergib University, Alkhums, Libya
Nafiz Arica Computer Engineering Department, Bahcesehir
University, Istanbul, Turkey
Zeinab Arjmandiasl Concordia Institute for Information Systems
Engineering (CIISE), Concordia University, Montreal, QC, Canada
Muhammad Azam Electrical and Computer Engineering (ECE),
Concordia University, Montreal, QC, Canada
B. Balasingam Department of Electrical and Computer
Engineering, University of Windsor, Windsor, ON, Canada
Jamal Bentahar Concordia Institute for Information Systems
Engineering (CIISE), Concordia University, Montreal, QC, Canada
F. Biondi Faculty of Human Kinetics, University of Windsor,
Windsor, ON, Canada
Aaron Boda Department of Earth and Space Science and
Engineering, Lassonde School of Engineering, York University,
Toronto, ON, Canada
Nizar Bouguila Concordia Institute for Information Systems
Engineering (CIISE), Concordia University, Montreal, QC, Canada
Duygu Cakir Computer Engineering Department, Bahcesehir
University, Istanbul, Turkey
Mark Green Faculty of Science, Ontario Tech University, Oshawa,
ON, Canada
Sorin Grigorescu
Robotics, Vision and Control (ROVIS), Transilvania University of
Brasov, Brasov, Romania
Artificial Intelligence, Elektrobit Automotive, Brasov, Romania
Baoxin Hu Department of Earth and Space Science and
Engineering, Lassonde School of Engineering, York University,
Toronto, ON, Canada
Xishi Huang RS Opto Tech Ltd., Suzhou, Jiangsu, China
M. Khalghollah Schulich School of Engineering, University of
Calgary, Calgary, AB, Canada
Howard Li Department of Electrical and Computer Engineering,
University of New Brunswick, Fredericton, NB, Canada
C.J.B. Macnab Schulich School of Engineering, University of
Calgary, Calgary, AB, Canada
Narges Manouchehri Concordia Institute for Information Systems
Engineering (CIISE), Concordia University, Montreal, QC, Canada
Jun Meng Zhejiang University Robotics Institute, Hangzhou,
China
Afshin Rahimi Department of Mechanical, Automotive and
Materials Engineering, University of Windsor, Windsor, ON, Canada
P. Ramakrishnan Department of Electrical and Computer
Engineering, University of Windsor, Windsor, ON, Canada
Jing Ren Department of Electrical and Computer Engineering,
Ontario Tech University, Oshawa, ON, Canada
Jianguo Wang Department of Earth and Space Science and
Engineering, Lassonde School of Engineering, York University,
Toronto, ON, Canada
Jin Wang Zhejiang University Robotics Institute, Hangzhou,
China
Chapter 1: A high-level design
process for neural-network
controls through a framework of
human personalities
M. Khalghollah; C.J.B. Macnab Schulich School of Engineering,
University of Calgary, Calgary, AB, Canada

Abstract
Current learning systems in the field of feedback control deal
with both unbiased nonlinearities and biased nonlinearities
(where bias is measured at the origin) quite differently.
Unbiased nonlinearities lend themselves to direct adaptive
control methods. Biased systems, on the other hand, typically
require actual learning of the bias term in order to achieve
acceptable error and effort. This paper attempts to unify these
approaches, and to learn to compensate for both types of
nonlinearities simultaneously. To do so we utilize a graphical,
quantitative theory of human personalities, which assumes that
their personalities indicate how people interact with the world
around them using feedback. This biologically-inspired
approach allows us to develop a formal design framework for
tackling this problem. Simulations with a two-link robotic
manipulator demonstrate the utility of the learning design
method, where gravity provides the main biased nonlinearities,
while friction, centripetal, and Coriolis forces are treated as
 unbiased nonlinearities; our neural-network update laws learn
all these robot nonlinearities at the same time. Lyapunov
methods result in stability guaranties for the proposed method.

Keywords
learning control; direct adaptive control; linear quadratic regulator;
gravity compensation; trajectory tracking; robotic manipulators;
personality theory
Chapter Outline

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
References

1.1 Introduction
This work addresses the difficulty in designing a learning control for
system dynamics that contain both a large bias and significant
nonlinearities, using the example of a robotic manipulator to
develop and test the ideas. The vast majority of learning-control
techniques in the literature addressing tracking control of the robot's
end-effector try to cancel the force of gravity, or compensate for
velocity-dependent nonlinearities that tend to pull the tip off-track,
i.e. friction, Coriolis, and centripetal (FCC) terms. The ad hoc
methods found in the literature that proposed learning both gravity
and FCC terms end up doing so sequentially, not simultaneously, to
the best of our knowledge. In this paper, we suggest the basic
difficulty in trying to achieve both at the same time stems from the
need to approach these two problems differently, i.e. they require
two different types of learning.
Gravity acting on a robot qualifies as a nonlinear term with
significant bias; for a simpler example of a biased nonlinearity,
consider a cosine term near the origin. Control systems generally
endeavor to cancel biased nonlinearities with a feedforward term, i.e.
an open-loop, bias-compensation, or set-point control. A sine
function, on the other hand, provides a simple example of a
nonlinear function with a value of zero at the origin; compensating
for this type of unbiased nonlinearity requires nonlinear feedback. A
learning control for this problem essentially ends up achieving a
memorized feedback term. Since the FCC terms in a robot manipulator
go to zero when the velocity goes to zero, and the need for high-
speed precision remains relatively rare in industrial applications, as
a practical matter the trajectory-tracking control problem without
gravity ends up more like learning sine than cosine. One important
difference from the designer's point of view lies in the choice of
robust weight update method, i.e. designing updates for the weights
(adaptive parameters) that remain robust to disturbances in the
sense of limiting weight drift (overlearning) and preventing bursting
(a sudden increase in error after a period of convergence). For
unbiased nonlinearities, a leakage term forgetting factor that tries to
drive the weights toward zero works well in practice. However, such
a term would directly result in a steady-state error for biased
nonlinearities; for biased nonlinearities the field of adaptive control
offers deadzone, parameter projection, and supervised leakage as
choices for the designer, all of which require some a priori
knowledge of the system's parameters and/or expected disturbances;
as a result many do not classify such methods as true learning
systems.
 Here we use our own previously-proposed graphical model of
human personalities to examine the problem [1]. A resulting
computational model, based on feedback theory, allows a prediction
of the probability distribution of Myers–Briggs behavioral technique
types in the human population, for the J/P and T/F pairs [2] as well
as for the I/E and N/S pairs [3]. This biologically-inspired perspective
allows more intuitive design methods for the high-level thinking
required in today's advanced control-system applications [4]. The
proposed framework provides insight into the nature of the two
types of learning problems outlined above, and it suggests how to
achieve them simultaneously on a robot arm, i.e. like humans do.
Building on the idea of LQR control, the approach results in an
optimization method for the design of all the control parameters
(feedback gains and adaptation rates) in a learning controller—the
first such formal design method appearing in the literature that
extends the LQR approach to nonlinear systems, to our knowledge.
Robotic-manipulator dynamics contain significant nonlinearities;
proposed control methods based on linear-system theory must, at
the very least, assume implementation of a gravity-compensation
term [5]. Actually learning the force of gravity should provide many
advantages, including decreasing engineering-design costs and
achieving real-time payload adaptation. An iterative learning control
[6] has some advantages, but does not address robustness to
disturbances and/or payloads. Radial basis function networks
(RBFNs) can learn to estimate the gravity term in some robot
manipulators [7,8] and some biped robots [9], but the method does
not extend well to multi-link robots due to the curse of dimensionality
when trying to add more inputs into an RBFN network.
The method in [10] learns both gravity and other nonlinearities,
but the very-small leakage term used in order to avoid sag-due-to-
gravity appears insufficient to deal with realistic disturbances. The
authors of [11] proposed an RBFN method that adapts to both
gravity (biased nonlinearity) and FCC terms (unbiased
nonlinearities), but requires knowledge of the inertia matrix—which
implies the designer would, in fact, know the gravity term. In [12],
only gravity compensation gets proposed in the first step of
 y g y p g p p p
backstepping for a flexible-joint robot, and not FCC terms. The same
authors tackle learning all nonlinearities for a Baxter robot in [13],
but the tracking performance requires a set of weights identified
during a learning stage—in practice it would seem the learning stage
would find the large biased nonlinearity and the tracking stage
would fine-tune the performance by compensating for the unbiased
nonlinearities.
In previous work, our research group presented a near-optimal
control [14], which developed an approach to achieve a near-optimal
control signal in the presence of gravity. A cerebellar model
articulation controller (CMAC) [15], with advantages of a fast
adaptation rate and real-time computational ability, was found to
have unique properties for tackling this problem; freezing a set of
supervisory network's local weights when the bias becomes
identified can compensate for the gravity bias term, and further
learning (using leakage) could then fine-tune for the FCC terms. The
disadvantage was the procedure was ad hoc, based on intuitive
insights into the workings of the CMAC.
Developing a formal design method, on the other hand, may have
some far-reaching implications for the field of control. Some have
playfully accused the field of control-systems of having a dirty secret:
designers often choose gains and/or parameters by trial-and-error,
rules-of-thumb, and/or experience. Such methods can be inadequate
when facing contemporary problems of interest, such as robots
interacting with unknown, unstructured environments.
Optimization methods promise a path to solve this problem, but
standard higher-level cognitive frameworks for the design of cost-
functionals remain an open problem to our knowledge. Even in the
case of the linear quadratic regulator (LQR), current theory does not
provide the designer with a method for choosing the Q and R
weighting matrices. For more advanced nonlinear systems that
interact with an environment, researchers struggle with even
creating a suitable cost-functional at the moment. This work
provides a biologically-inspired method for designing cost-
functionals and the value of the weightings. We use a model of
human personalities; choosing a personality directly results in a
 p g p y y
choice of numerical weightings. Thus, a control system designer can
use their intuitive understanding of human personalities at the high-
level design stage. Not only can this avoid trial and error in picking
parameters, but it may also significantly reduce the total number of
parameters needed compared to manual designs. Consider an
analogy to how fuzzy logic proved quite a time-saving invention for
control design, as a result of allowing human intuition to guide the
design of computational decision-making and reducing the number
of parameters that the designer needs to choose; fuzzy control ended
up significantly broadening the field of computer automation, since
many more problems would lend themselves to a cost-effective
and/or time-efficient design solution.
This chapter first gives a background on CMAC neural network
and direct adaptive control methods, as applied to both biased and
unbiased nonlinearities. The Background section ends with a short
introduction to our personality theory, and describes how the
framework enables design of PID+bias controls using a nonlinear
quadratic regulator. In the Proposed Methods section, we show how
to extend the approach to designing an adaptive learning control.
The Results section illustrates a formal engineering design
procedure, based on constraints and objectives, for a simple mass-
spring-mass and then a two-link robotic manipulator.

1.2 Background
1.2.1 The CMAC associative-memory neural
network
This work assumes an associative memory will serve as the
nonlinear approximator without loss of generality, where a weighted
sum of basis functions gives the estimate of nonlinear function
as
 (1.1)

where contains the n inputs, the weights reside in column

vector , and denotes a row vector of basis functions.
Typically one uses Gaussians as the basis functions in a radial basis
function network (RBFN). Since the RBFN's curse of dimensionality
makes it difficult to use for more than three inputs, the learning
controls in this paper use a Cerebellar Model Articulation Controller
(CMAC) instead. The CMAC consists of hypercube-domain basis
functions, or cells, constructed with rectangles, triangles, or splines.
The local nature of the CMAC cells ensures only a small number of
activated cells require calculations, along with their corresponding
weights, during any one time-step of the control.
One could build a CMAC using m offset layers of n-dimensional
arrays. The input activates (indexes) one cell per layer. The
construction of the basis functions in a spline-CMAC follows

(1.2)
where denotes the normalized position within the
cell on the kth input; normalization produces

(1.3)
An example of a spline CMAC with normalized basis functions in
one dimension appears in Fig. 1.1: for two dimensions see Fig. 1.2.
Rather than an impractical allocation of memory for n-dimensional
arrays, the CMAC uses hash-coding to map virtual cells to a one-
dimensional physical memory, since relatively few cells ever get
activated in high-dimensional space during trajectory tracking of a
robot [16] (hash collisions become possible, however unlikely).

FIGURE 1.1 One-input CMAC with spline membership

functions: m = 3 layers and Q = 10 discretizations. The
output is a weighted sum of basis functions.
 FIGURE 1.2 Normalized-spline, two-input CMAC: m = 4 and
Q = 4. The word layers for a CMAC refers to hash-coded
arrays indexed in parallel.

Being a universal nonlinear approximator, the CMAC can

approximate nonlinear function as

(1.4)

where defines the approximation error in the uniform

approximation region : specifically for a
positive constant .

1.2.2 Unbiased nonlinearities

In this work the learning control strategies come from mathematical
techniques developed in the field of direct adaptive control.
Consider a physical system with position x and control input u

(1.5)

where the unbiased nonlinearity has , with constant

parameters , , and . Given desired trajectory ,
then defining state errors , and auxiliary error
leads to an expression of the dynamics suitable for
control-system design

(1.6)

Although a desired trajectory may introduce a system bias, such that

, a practical neural-adaptive control ignores this typically-
small effect and a common design uses leakage [17] multiplied by
positive parameter ν in the robust weight update

(1.7)

(1.8)
where indicates the ideal-weight estimates, positive parameter ν
determines the influence of the robust leakage modification, which
limits weight drift by trying to force the weights to zero. As long as
remains relatively small, the leakage term does not significantly
reduce performance. The definition of z implies positive constant
equates to a derivative gain from PID control—giving an effective
proportional gain . We denote the positive-constant
adaptation gain as because in subsequent sections will we point
out similarities of the neural network to an integral term, i.e. this
work treats a neural network trained with adaptive-control update
laws as just a memorized integral.
Note that the system control actually occurs as a multi-rate signal,
with the feedback running fast enough to approximate a continuous
time signal and the neural network updated at a discrete rate, i.e. a
zero-order-hold discretized signal describes its output characteristics.
Designers often choose leakage for a robust update, and in discrete
time with sampling period Δt the delta-rule update becomes

(1.9)

Note the delta-rule simply results in the neural network estimating

the discrete-time model rather than the continuous-time model using
its universal approximation ability, i.e. it does not cause instability
when Δt is relatively small [18].

1.2.3 Direct adaptive control in the presence of

bias
Consider adding a biased nonlinearity, with , to the
previous model
 (1.10)

In this case the leakage term multiplied by ν in (1.8) typically pushes

the weights in the wrong direction and performance suffers;
deadzone provides a more practical robust modification

(1.11)

where includes the maximum approximation error of the neural

network within its domain , combined with maximum external-
disturbance amplitude. Thus, the amount of system-plus-
disturbance knowledge required may make the method impractical
in many applications.
Many papers in the literature suggest adaptive-parameter
projection instead, even simple versions like

(1.12)

The designer requires knowledge of the maximum and minimum

weight for the i cell, and respectively, presumably
identified during pre-training.
 Without a priori knowledge of the system and disturbances,
learning control becomes the logical choice in the context of biased
nonlinearities. A direct adaptive control using supervised leakage,

(1.13)

can incorporate a learning control scheme when the weights that

estimate the bias are output from an online algorithm attempting
to identify the bias in real time; for stability reasons should only
be updated once. The robust term contains multiplying parameter
denoted because in subsequent sections will show this term
has similar effects as a control-bias term (a setpoint or feedforward
term).

1.2.4 A graphical model of personalities

The graphical model of personalities consists of a quadrant with
empowerment/manipulation on the horizontal axis, indicating how
one tries to affect the world and people, and emotion/logic on the
vertical axis indicating how one makes decisions. Manipulation
includes commands, requests, charm, and all subtle forms of the art.
Empowerment includes running away, avoiding, giving away power
—and at higher levels of thinking includes creating safe structures,
teaching, learning, planning, among many other things that do not
clearly come across as manipulation. The vertical axis denotes the
emotion and logic. Emotion provides coded memories, felt rather
than remembered, for quick access and for giving useful direction.
By logic we mean deductive logic, based on a set of beliefs; recall logic
qualifies as either valid or invalid, whereas beliefs classify as either
consistent or inconsistent. For instance, two people can disagree all
while following perfectly logical arguments based on their own
beliefs. Also, people only have a (perhaps delusional) self-image of
performing valid logical deductions and predictions.
Eight of the clinical personality disorders (PDs) anchor the system,
since they exhibit widely-agreed-upon qualities. Histrionic and
Borderline lie in the upper-right (UR), Avoidant and Dependent in
the upper-left (UL), Schizoid and Schizotypal sit in the lower-left
(LL), and finally the lower-right (LR) has Narcissistic and Antisocial.
The characteristic on the axis to the counter-clockwise for each
personality disorder seems quite apparent to all as the main
behavior; everyone experiences URs as emotional, ULs are seen to
empower others by leaving them alone or surrendering all their
power, LLs are seem overly logical when they are creating theories
(even if the underlying belief system seems unique), and LRs have
reputations for their shameless manipulation. The hidden
characteristic becomes apparent as a main motivator typically only
to people who engage in intimate relationships with PDs. Inside the
circle indicated in Fig. 1.6, better-balanced people have personality
self-images rather than PDs; for personalities the visible and hidden
characteristics predict biases rather than behaviors. The visible axis
indicates a projected bias, felt as an irrational hope; we describe the
hidden as a reflected bias, felt as an irrational fear of others exhibiting
this characteristic. Note that one does not necessarily move outwards
on the graph to a PD in one's quadrant if one was having mental
health issues [19], [20], and our model does not yet predict
movement on the graph. Personality theories that indicate self-
images include Lowry Colors and the Enneagram.
The Lowry True Colors system [21] provides personality
descriptions that closely match the mentioned qualities (Fig. 1.6).
Briefly, the Green personality types feel primarily logical; many
engineering undergraduate students feel Green. The Gold
personality types feel mission-oriented and like to empower others
through sharing knowledge and meticulous planning; many
teachers experience themselves as Gold. The Orange personality
types feel action-oriented and like to get things done by leading
(manipulating) others; many managers perceive themselves as
Orange. The Blue personality types feel primarily people-oriented
 g p y yp p y p p
and focus on building relationships using their emotions; many
empathy-based therapists experience themselves as blue.
The Enneagram has found vast popularity with the general public
and has nine types. If we move the types 1, 4, and 8 into new
positions it achieves consistency with the quadrant system. Full
qualitative descriptions of how the personality disorders, Lowry
Colors, and Enneagram personalities fit on the self-image quadrant
appear in [1]. Also by mapping Merrill's C.A.P.S. theory of work
place personalities [22] onto the quadrant, when considering it as the
way others perceive someone. For example, opposite types in the
Enneagram come across similarly to the outside world because
opposite self-images also imply opposite unconscious and opposite
unconscious communications.
The quadrant matches closely to how people in Western society
anthropomorphize some animals; a biological explanation for
personalities apparently becomes a possibility if we imagine that,
earlier in evolution, empowerment may have started with just passive
behavior. Manipulation may have started with simple active (Fig. 1.5)
movements. What if we consider animals as a type of robot,
engaging with the world through feedback? Then active may
indicate closed-loop control, and passive open-loop control. For a
human example, think of an athlete learning a new movement; at
first it is an active movement that requires closed-loop control
relying on feedback, with resulting excessive effort. Constant
practice and training will eventually turn the movement into a
memorized passive movement using open-loop control, resulting in
little (minimal) effort due to relaxed muscles.

1.2.5 A computer model of personalities

When the sensors provide measurements of the system error e and
its derivative , most industrial applications use a traditional
PID+bias control
 (1.14)

where the constants , , and defines the control gains, and

defines the control bias. In order to design optimal gain parameters,
one must choose a cost-functional. The reader may already know
how to use an LQR cost-functional to pick optimal PD gains; here we
use a nonlinear quadratic regulator cost functional based on a
framework of personality self-images

(1.15)
where the words indicate overall norm measurements e.g.
norms; the first two terms equate to the terms multiplied by the Q
matrix in an LQR control (assuming an SISO system with position
and velocity states), while the third term penalizes the PID control
effort for the first half of the step response, but for the second half
acts like traditional LQR only penalizing a measure of PD feedback
effort, e.g. . The term stems from understanding the
integral term as trying to compensate for the bias (or inaccuracies in
bias compensation). Thus, an integral qualifies as a simple learning
term, and its error measures the distance from its ideal value (the
bias that would result in as the error reaches zero at steady
state). We point out the following similarities to the qualities in the
graphical theory of personalities:
which allows a high-level control design e.g. choosing a desired
personality results in the weighting parameters in a nonlinear
quadratic regulator (NQR). This eliminates the need to pursue trial-
and-error, rule-of-thumb, or experiential choices of Q's and R's,
allowing an intuitive understanding of human personalities to guide
the design process at the highest level (Fig. 1.7).
In previous work we pointed out similarities in PID+bias control
to the Myers–Briggs personality behavioral techniques for humans:

and this model (Fig. 1.3) allowed us to predict the probability

distribution of Myers–Briggs conflict types (J/P and T/F) in the
human population. The optimization problem only requires three
parameters, so the four-parameter PID+bias is overparameterized; in
terms of human personalities we view this as free will for an
individual. Thus, the optimization must start with one a priori
parameter which we will refer to as our free-will parameter.
 FIGURE 1.3 Feedback model used to explain the
probability distribution of J/P and T/F in the human
population in [2].

The designer can first choose a personality, a precise angle on the

personality quadrant, and then choose a personality imbalance for a
magnitude. Consider using a parameter p to define the ratio of
personality characteristics on opposite sides of the graph: for
example in Fig. 1.8 the designer has chosen a Type 5 personality and
by choosing and the other qualities stem from
the chosen imbalance, and with .
Consider the proposed precise design for the cost-functional

(1.16)

where and defines the learning error of

the integral term. The integral effort gets penalized during the first
half of the step response, but after that we worry only about its
learning error. PD control defines the pure feedback component
while the memory (integral) term uses
initial condition . The algorithm sets an initial condition for
the integral term as in the first run and then at for the
second. Note the first term simply uses traditional LQR design; the
NQR simply adds a component that encourages the integral's
learning error to draw near zero by time .
Thus, a nonlinear quadratic regulator producing an optimal
PID+bias control models how humans choose a behavioral technique
(Myers–Briggs) to try and achieve consistency with their self-image
personality (Enneagram); since the model stems from the principles
of feedback, the assumption that evolution did not find a way to
escape the principles of feedback implies this model may indeed
capture the underlying principles of human personalities. Modeling
the other Myers–Briggs letters with feedback (Fig. 1.4) allowed us to
predict the probability distribution of N/S (given I/E) in the human
population, and the interested reader might refer to [23] for more
detailed explanations.

FIGURE 1.4 Control-system model from [3] used to predict

I/E and N/S distributions.
FIGURE 1.5 Personality self-images of Lowry True Colors
can be understood by anthropomorphizing animals. Green
ellipse: many engineering students may identify with Orcas
who learn logical hunting techniques, which may be called
passive since they are very careful about what they eat.

FIGURE 1.6 The clinical personality disorders represent

unbalanced people. Green ellipse: a shizoid tells people
about their theories (visible logic), with the idea of warning,
helping, or enlightening others (hidden empowerment).
 FIGURE 1.7 The proposed new Enneagram configuration
moves types 1, 4, and 8 to make it consistent with the
quadrant system. Noticing similarities between the four
personality qualities and Q1, Q2, R1, and R2 in an NQR
control results in an intuitive control-system design
process. The oval gives way to visualize the self-image of
a Type 5, for example, who has all four qualities but not in
equal measure.

1.3 Proposed methods

1.3.1 Proposed learning law
Based on our previous observations we propose using supervised
leakage

(1.17)

where the bias weights start at zero, , and

 (1.18)

where the moment in time the algorithm captures a bias estimate,

i.e. when the system first begins to remain close to the origin with
close to zero velocity or, equivalently, a settling time. We only expect
to capture an estimate of the bias, and the continued updates make
up for inaccuracies in this bias estimation and compensate for more
detailed characteristics of the nonlinearities near the origin.

1.3.2 Cost functional for optimization

Using the similarity between the four quadrant axes and the
standard LQR problem, we propose the cost functional

(1.19)

The first three terms look familiar, known from the LQR cost
functional, while the feedback effort only includes

(1.20)

and not the neural network outputs (Fig. 1.9).

 The total extra effort should penalize the effort needed above a
positive bias, but not any effort below the bias. Here we look at this
term as an opportunity to also limit weight drift; rather than
measure neural network output directly we look at the level of
individual weights, and penalize each weight for being too large-
positive. Otherwise, a weight activated only in a region where the
basis function had a small value wouldn't get penalized enough for
its growth; the proposed method ensures each weight gets equally
penalized for growing regardless of the value of the basis function.
This measurement of extra effort gets updated at time step k, of
length Δt, according to

The learning error denotes the difference between neural network

portion of the control and all the nonlinear terms; describing by
functions of time gives the expression

Unlike an LQR control, the design with an integral must consider

initial conditions. Thus, we run the step response twice; the first has
all weights initialized to zero and the second keeps the same weight
values from the end of the first run. Thus, the total cost functional
becomes
 (1.21)
with initialcondition for the two step responses
, where and ; the simulation in the
Results section uses .

1.3.3 Stability analysis

This section investigates stability theory in terms of Lyapunov
functions. The optimization values , , and will dive form
the personality angle and magnitude; themselves functions of the
personality imbalance parameters (see Fig. 1.8 for an example). The
imbalance in the personality gets described by parameter p. Using
the Matlab® optimization tool fmincon() generates the optimal
values for , and with the proportional gain as input, i.e. a free
will parameter.

FIGURE 1.8 Example: choice of Type 5 personality with

imbalance parameter p = 0.5 results in particular values for
the NQR cost-functional weightings.
 FIGURE 1.9 Computational model stemming from Fig. 1.3.
NQR models Enneagram and PID+bias models MB
conflict types (J/P and T/F).

A Lyapunov candidate Eq. (1.22) enables stability boundary

identification when the designed control law (1.7) and the proposed
update term (1.13) gets applied to (1.5). The adaptive Lyapunov
functions

(1.22)

where indicates the weight errors, with column vector w

indicating the (unknown) ideal values of the weights, results in time
derivative

(1.23)

Those familiar with (supervised) leakage might recall that (1.23)

implies signals are uniformly ultimately bounded (see the
appendix).
1.4 Results
1.4.1 Developing a design procedure
Consider the vertical mass–spring–mass system

(1.24)

(1.25)

where y and x describe the positions of the two masses and

, , denote friction coefficients, gives the
spring constant, describes gravitational acceleration, and u
provides the input force. Note so that the error
ensures regulation about the position , i.e. gravity forms part of
the biased nonlinearity. The second equation represents unmodeled
dynamics. The nonlinearity, including both biased and unbiased
terms, consists of

(1.26)

This method proposed a formal engineering design process for

learning controls based on meeting objectives and constraints—
perhaps the first of its kind ever proposed. We use the following
constraints in the design, monitored during a step response:
 Constraint 1) Position constraint m,
Constraint 2) Velocity constraint m/s,
Constraint 3) Effort constraint bias.

Thus, the position constraint defines what the user would see as the
effective steady-state error for the step response, the remaining sag
below the set-point at the end of the measured step. The velocity
constraint might help ensure safe operation and/or help reduce
wear-and-tear of mechanical parts e.g. so that the system lasts the
specified life-span as per the engineering specs. One might design
the effort constraints with actuator saturation, actuator life-span,
and/or safety considerations in mind. For objectives we look at
minimizing total error, velocity, and/or effort over the entire step.

Objective 1) Performance objective is minimizing ,

Objective 2) Velocity objective is minimizing ,
Objective 3) Effort objective is minimizing
.

The design of the saturation function assumes a gravity down

scenario, so that extra effort only occurs when pushing more than
the force of gravity

(1.27)

so that the effort objective measures both feedback effort and extra
effort (beyond the bias).
We propose the following engineering design procedure:
 Step 1) Start with very unbalanced personalities—choose
parameter p small.
Step 2) Run step responses for all eight personality types.
Step 3) Repeat first two steps until at least three personalities
meet all the constraints.
Step 4) Compare how well the constraint-satisfying controls
meet the objectives.
Step 5) Choose one of these controls.

We started with the imbalance parameter . The imbalance

parameter determined as a multiplier for determining the amount of
personality offset. All simulations use the proportional gain as the
free will parameter and set for every optimization. Each
optimization starts from five different random initial conditions in
an attempt to find a minimum close to (or exactly on) the global
minimum, for the remaining parameters .
The algorithm to identify the bias used in the supervisory leakage
term (1.17) was: if both the magnitude of the position error remained
below 0.1 m and velocity magnitude remained below 0.1 m for 10
discrete time steps the supervisory weights were set to the current
value of the weight estimates .
A first attempt using reveals no personalities meet the
constraints (Fig. 1.10). But a more balanced personality with
(Fig. 1.11) has four personalities that meet the constraints; the
objectives (Fig. 1.12) reveal Type 8 does best for velocity (Fig. 1.13),
Type 6 shows the best result for position performance by a very
slight amount (Fig. 1.14), Type 7 shows the best result for effort and
does not appear to sacrifice performance to do so (Fig. 1.15), while
Type 3 does not do particularly well at anything (Fig. 1.16). Thus, it
seems likely a designer would choose Type 7 based on these results,
which has optimization result in , , and . Note
these figures also compare the proposed adaptive-NQR performance
to a standard PID+bias design; the PD gains came from an LQR
optimization, with the bias 0.9 of the true bias modeling (which
could represent an unknown payload), and the integral gain came
from trial-and-error, i.e. like one might do in industrial application—
the result was , , , and .

FIGURE 1.10 Normalized constraints with p = 0.1: inside

the circle constraints are met. No personality met all three
constraints.

FIGURE 1.11 Increasing balance by using p = 0.5. Type 7,

6, 8 and also 3 personalities now meet all three
constraints.
FIGURE 1.12 Normalized objectives with p = 0.5: closer to
the middle is better. Of those that meet constraints, all four
have similar position performance but Type 7 does best
minimizing control effort.

FIGURE 1.13 Type 8: does well on both position constraint

and position objective.
Other documents randomly have
different content
"No, sir," said the boy. "He's in hospital, shot."

"Is there a key to his place? Quick."

"I think so, sir, but I can't give it to you."

"Here, give it to me, then!" exclaimed the detective. He flashed a

badge in the boy's eyes, and the youth immediately lost a deal of his
coolness.

"Gee, a detective! Yes, sir."

"How many rooms has Mr. St. George?" asked the scientist.

"Three and a bath," the boy responded.

Two minutes later the three men stood in the reception-room of the
apartments. There came to them from somewhere inside a deadly,
stifling odor of chloroform. After one glance around The Thinking
Machine rushed into the next room, the studio.

"Dear me, dear me!" he exclaimed.

There on the floor lay huddled the figure of a man. Blood had run
from several wounds on his head. The Thinking Machine stooped a
moment, and his slender fingers fumbled over the heart.

"Unconscious, that's all," he said, and he raised the man up.

"Victor Willis!" exclaimed Hatch.

"Victor Willis!" repeated The Thinking Machine, as if puzzled. "Are

you sure?"

"Certain," said Hatch, positively. "It's the bank clerk."

"Then we are too late," declared the scientist.

He arose and looked about the room. A door to his right attracted
his attention. He jerked it open and peered in. It was a clothes
press. Another small door on the other side of the room was also
thrown open. Here was a kitchenette, with a great quantity of
canned stuffs.

The Thinking Machine went on into the little bedroom which Hatch
had searched. He flung open the bathroom and peered in, only to
shut it immediately. Then he tried the handle of another door, a
closet. It was fastened.

"Ah!" he exclaimed.

Then on his hands and knees he sniffed at the crack between the
door and the flooring. Suddenly, as if satisfied, he arose and stepped
away from the door.

"Smash that door in," he directed.

Detective Mallory looked at him stupefied. There was a similar

expression on Hatch's face.

"'What's--what's in there?" the detective asked. "Smash it," said the

other, tartly. "Smash it, or God knows what you'll find in there."

The detective, a powerful man, and Hatch threw their weight against
the door; it stood rigid. They pulled at the handle; it refused to yield.

"Lend me your revolver?" asked The Thinking Machine.

The weapon was in his hand almost before the detective was aware
of it, and, placing the barrel to the keyhole, The Thinking Machine
pulled the trigger. There was a resonant report, the lock was
smashed and the detective put out his hand to open the door.

"Look out for a shot," warned The Thinking Machine, sharply.

 VI.

The Thinking Machine drew Detective Mallory and Hatch to one side,
out of immediate range of any person who might rush out, then
pulled the closet door open. A cloud of suffocating fumes--the
sweet, sickening odor of chloroform--gushed out, but there was no
sound from inside. The detective looked at The Thinking Machine
inquiringly.

Carefully, almost gingerly, the scientist peered around the edge of

the door. What he saw did not startle him, because it was what he
expected. It was Constans St. George lying prone on the floor as if
dead, with a blood-spattered revolver clasped loosely in one hand;
the other hand grasped the throat of a woman, a woman of superb
physical beauty, who also lay with face upturned, staring glassily.

"Open the windows--all of them, then help me," commanded the

scientist.

As Detective Mallory and Hatch turned to obey the instructions, The

Thinking Machine took the revolver from the inert fingers of the
artist. Then Hatch and Mallory returned and together they lifted the
unconscious forms toward a window.

"It's Grace Field," said the reporter.

In silence for half an hour the scientist labored over the unconscious
forms of his three patients. The detective and reporter stood by,
doing only what they were told to do. The wind, cold and stinging,
came pouring through the windows, and it was only a few minutes
until the chloroform odor was dissipated. The first of the three
unconscious ones to show any sign of returning comprehension was
Victor Willis, whose presence at all in the apartments furnished one
of the mysteries which Hatch could not fathom.

It was evident that his condition was primarily due to the wounds on
his head--two of which bled profusely. The chloroform had merely
served to further deaden his mentality. The wounds were made with
the butt of the revolver, evidently in the hands of the artist. Willis's
eyes opened finally and he stared at the faces bending over him with
uncomprehending eyes.

"What happened?" he asked.

"You're all right now," was the scientist's assuring answer. "This man
is your prisoner, Detective Mallory, for breaking and entering and for
the attempted murder of Mr. St. George."

Detective Mallory was delighted. Here was something he could

readily understand; a human being given over to his care; a tangible
thing to put handcuffs on and hold. He immediately proceeded to
put the handcuffs on.

"Any need of an ambulance?" he asked.

"No," replied The Thinking Machine. "He'll be all right in half an

hour."

Gradually as reason came back Willis remembered. He turned his

head at last and saw the inert bodies of St. George and Grace Field,
the girl whom he had loved.

"She was here, then!" he exclaimed suddenly, violently. "I knew it. Is
she dead?"

"Shut up that young fool's mouth, Mr. Mallory," commanded the

scientist, sharply. "Take him in the other room or send him away."
Obediently Mallory did as directed; there was that in the voice of this
cold, calm being, The Thinking Machine, which compelled obedience.
Mallory never questioned motives or orders.

Willis was able to walk to the other room with help. Miss Field and
St. George lay side by side in the cold wind from the open window.
The Thinking Machine had forced a little whiskey down their throats,
and after a time St. George opened his eyes.

The artist was instantly alert and tried to rise. He was weak,
however, and even a strength given to him by the madness which
blazed in his eyes did not avail. At last he lay raving, cursing,
shrieking. The Thinking Machine regarded him closely.

"Hopeless," he said, at last.

Again for many minutes the scientist worked with the girl. Finally he
asked that an ambulance be sent for. The detective called up the
City Hospital on the telephone in the apartments and made the
request. The Thinking Machine stared alternately at the girl and at
the artist.

"Hopeless," he said again. "St. George, I mean."

"Will the girl recover?" asked Hatch.

"I don't know," was the frank reply. "She's been partly stupefied for
days--ever since she disappeared, as a matter of fact. If her physical
condition was as good as her appearance indicates she may recover.
Now the hospital is the best place for her."

It was only a few minutes before two ambulances came and the
three persons were taken away; Willis a prisoner, and a sullen,
defiant prisoner, who refused to speak or answer questions; St.
George raving hideously and cursing frightfully; the woman,
beautiful as a marble statue, and colorless as death.
When they had all gone, The Thinking Machine went back into the
bedroom and examined more carefully the little closet in which he
had found the artist and Grace Field. It was practically a padded cell,
relatively six feet each way. Heavy cushions of felt two or three
inches thick covered the interior of the little room closely. In the top
of it there was a small aperture, which had permitted some of the
fumes of the chloroform to escape. The place was saturated with the
poison.

"Let's go," he said, finally.

Detective Mallory and Hatch followed him out and a few minutes
later sat opposite him in his little laboratory. Hatch had told a story
over the telephone that made his City Editor rejoice madly; it was
news, great, big, vital news.

"Now, Mr. Hatch, I suppose you want some details," said The
Thinking Machine, as he relapsed into his accustomed attitude. "And
you, too, Mr. Mallory, since you are holding Willis a prisoner on my
say-so. Would you like to know why?"

"Sure," said the detective.

"Let's go back a little--begin at the beginning, where Mr. Hatch called

on me," said The Thinking Machine. "I can make the matter clearer
that way. And I believe the cause of justice, Mr. Mallory, requires
absolute accuracy and clarity in all things, does it not?"

"Sure," said the detective again.

"Well, Mr. Hatch told me at some length of the preliminaries of this

case," explained The Thinking Machine. "He told me the history of
the picture; the mystery as to the identity of the model; her great
beauty; how he found her to be Grace Field, a shop-girl. He also told
me of the mental condition of the artist, St. George, and repeated
the rumor as he knew it about the artist being heartbroken because
the girl--his model--would not marry him.
"All this brought the artist into the matter of the girl's disappearance.
She represented to him, physically, the highest ideal of which he
could conceive--hope, success, life itself. Therefore it was not
astonishing that he should fall in love with her; and it is not difficult
to imagine that the girl did not fall in love with him. She is a
beautiful woman, but not necessarily a woman of mentality; he is a
great artist, eccentric, childish even in certain things. They were two
natures totally opposed.

"These things I could see instantly. Mr. Hatch showed me the

photograph and also the scrap of paper. At the time the scrap of
paper meant nothing. As I pointed out, it might have no bearing at
all, yet it made it necessary for me to know whose handwriting it
was. If Willis's, it still might mean nothing; if St. George's, a great
deal, because it showed a direct thread to him. There was reason to
believe that any friendship between them had ended when the
picture was exhibited.

"It was necessary, therefore, even that early in the work of reducing
the mystery to logic to center it about St. George. This I explained to
Mr. Hatch and pointed out the fact that the girl and the artist might
have eloped--were possibly together somewhere. First it was
necessary to get to the artist; Mr. Hatch had not been able to do so.

"A childishly simple trick, which seemed to amaze Mr. Hatch

considerably, brought the artist out of his rooms after he had been
there closely for two days. I told Mr. Hatch that the artist would
leave his rooms, if he were there, one night at 9:32, and told him to
wait in the hall, then if he left the door open to enter the apartments
and search for some trace of the girl. Mr. St. George did leave his
apartments at the time I mentioned, and----"

"But why, how?" asked Hatch.

"There was one thing in the world that St. George loved with all his
heart," explained the scientist. "That was his picture. Every act of his
life has demonstrated that. I looked at a telephone book; I found he
had a 'phone. If he were in his rooms, locked in, it was a bit of
common sense that his telephone was the best means of reaching
him. He answered the 'phone; I told him, just at 9:30, that the Art
Museum was on fire and his picture in danger.

"St. George left his apartments to go and see, just as I knew he

would, hatless and coatless, and leaving the door open. Mr. Hatch
went inside and found two gloves and a veil, all belonging to Miss
Field. Miss Stanford identified them and asked if he had gotten them
from Willis, and if Willis had been arrested. Why did she ask these
questions? Obviously because she knew, or thought she knew, that
Willis had some connection with the affair.

"Mr. Hatch detailed all his discoveries and the conversation with Miss
Stanford to me on the day after I 'phoned to St. George, who, of
course, had found no fire. It showed that Miss Stanford suspected
Willis, whom she loved, of the murder of Miss Field. Why? Because
she had heard him threaten. He's a hare-brained young fool,
anyway. What motive? Jealousy. Jealousy of what? He knew in some
way that she had posed for a semi-nude picture, and that the man
who painted it loved her. There is your jealousy. It explains Willis's
every act."

The Thinking Machine paused a moment, then went on:

"This conversation with Mr. Hatch made me believe Miss Stanford

knew more than she was willing to tell. In what way? By a letter?
Possibly. She had given Mr. Hatch a scrap of a letter; perhaps she
had found another letter, or more of this one. I sent her a note,
telling her I knew she had these scraps of letters, and she promptly
brought them to me. She had found them after Mr. Hatch saw her
first somewhere in the house--in a bureau drawer she said, I think.

"Meanwhile, Mr. Hatch had called my attention to the burglary of St.

George's apartments. One reading of that convinced me that it was
Willis who did this. Why? Because burglars don't burst in doors when
they think anyone is inside; they pick the lock. Knowing, too, Willis's
insane jealousy, I figured that he would be the type of man who
would go there to kill St. George if he could, particularly if he
thought the girl was there.

"Thus it happened that I was not the only one to think that St.
George knew where the girl was. Willis, the one most interested,
thought she was there. I questioned Miss Stanford mercilessly, trying
to get more facts about the young man from her which would bear
on this, trying to trick her into some statement, but she was loyal to
the last.

"All these things indicated several things. First, that Willis didn't
actually know where the girl was, as he would have known had he
killed her; second, that if she had disappeared with a man, it was St.
George, as there was no other apparent possibility; third, that St.
George would be with her or near her, even if he had killed her;
fourth, the pistol shot through the arm had brought on again a
mental condition which threatened his entire future, and now as it
happens has blighted it.

"Thus, Miss Field and St. George were together. She loved Willis
devotedly, therefore she was with St. George against her will, or she
was dead. Where? In his rooms? Possibly. I determined to search
there. I had just reached this determination when I heard St.
George, violently insane, had escaped from the hospital. He had only
one purpose then--to get to the woman. Then she was in danger.

"I reasoned along these lines, rushed to the artist's apartments,

found Willis there wounded. He had evidently been there searching
when St. George returned, and St. George had attacked him, as a
madman will, and with the greater strength of a madman. Then I
knew the madman's first step. It would be the end of everything for
him; therefore the death of the girl and his own. How? By poison
preferably, because he would not shoot her--he loved beauty too
much. Where? Possibly in the place where she had been all along,
the closet, carefully padded and prepared to withstand noises. It is
really a padded cell. I have an idea that the artist, sometimes
overcome by his insane fits, and knowing when they would come,
prepared this closet and used it himself occasionally. Here the girl
could have been kept and her shrieks would never have been heard.
You know the rest."

The Thinking Machine stopped and arose, as if to end the matter.

The others arose, too.

"I took you, Mr. Mallory, because you were a detective, and I knew I
could force a way into the apartments which I imagined would be
locked. I think that's all."

"But how did the girl get there?" asked Hatch.

"St. George evidently asked her to come, possibly to pose again. It

was a gratification to the girl to do this--a little touch of vanity
caused her to pose in the first place. It was this vanity that Willis
was fighting so hard, and which led to his threats and his efforts to
kill St. George. Of course the artist was insane when she came; his
frantic love for her led him to make her a prisoner and hold her
against her will. You saw how well he did it."

There was an awed pause. Hatch was rubbing the nap of his hat
against his sleeve, thoughtfully. Detective Mallory had nothing to
say; it was all said. Both turned as if to go, but the reporter had two
more questions.

"I suppose St. George's case is hopeless?"

"Absolutely. It will end in a few months with his death."

"And Miss Field?"

"If she is not dead by this time she will recover. Wait a minute." He
went into the next room and they heard the telephone bell jingle.
After a time he came out. "She will recover," he said. "Good-
afternoon."

Wonderingly, Hutchinson Hatch, reporter, and Detective Mallory

passed down the street together.

THE END
 *** END OF THE PROJECT GUTENBERG EBOOK THE PROBLEM OF
CELL 13 ***

Updated editions will replace the previous one—the old editions will
be renamed.

Creating the works from print editions not protected by U.S.

copyright law means that no one owns a United States copyright in
these works, so the Foundation (and you!) can copy and distribute it
in the United States without permission and without paying
copyright royalties. Special rules, set forth in the General Terms of
Use part of this license, apply to copying and distributing Project
Gutenberg™ electronic works to protect the PROJECT GUTENBERG™
concept and trademark. Project Gutenberg is a registered trademark,
and may not be used if you charge for an eBook, except by following
the terms of the trademark license, including paying royalties for use
of the Project Gutenberg trademark. If you do not charge anything
for copies of this eBook, complying with the trademark license is
very easy. You may use this eBook for nearly any purpose such as
creation of derivative works, reports, performances and research.
Project Gutenberg eBooks may be modified and printed and given
away—you may do practically ANYTHING in the United States with
eBooks not protected by U.S. copyright law. Redistribution is subject
to the trademark license, especially commercial redistribution.

START: FULL LICENSE

THE FULL PROJECT GUTENBERG LICENSE
 PLEASE READ THIS BEFORE YOU DISTRIBUTE OR USE THIS WORK

To protect the Project Gutenberg™ mission of promoting the free

distribution of electronic works, by using or distributing this work (or
any other work associated in any way with the phrase “Project
Gutenberg”), you agree to comply with all the terms of the Full
Project Gutenberg™ License available with this file or online at
www.gutenberg.org/license.

Section 1. General Terms of Use and

Redistributing Project Gutenberg™
electronic works
1.A. By reading or using any part of this Project Gutenberg™
electronic work, you indicate that you have read, understand, agree
to and accept all the terms of this license and intellectual property
(trademark/copyright) agreement. If you do not agree to abide by all
the terms of this agreement, you must cease using and return or
destroy all copies of Project Gutenberg™ electronic works in your
possession. If you paid a fee for obtaining a copy of or access to a
Project Gutenberg™ electronic work and you do not agree to be
bound by the terms of this agreement, you may obtain a refund
from the person or entity to whom you paid the fee as set forth in
paragraph 1.E.8.

1.B. “Project Gutenberg” is a registered trademark. It may only be

used on or associated in any way with an electronic work by people
who agree to be bound by the terms of this agreement. There are a
few things that you can do with most Project Gutenberg™ electronic
works even without complying with the full terms of this agreement.
See paragraph 1.C below. There are a lot of things you can do with
Project Gutenberg™ electronic works if you follow the terms of this
agreement and help preserve free future access to Project
Gutenberg™ electronic works. See paragraph 1.E below.
1.C. The Project Gutenberg Literary Archive Foundation (“the
Foundation” or PGLAF), owns a compilation copyright in the
collection of Project Gutenberg™ electronic works. Nearly all the
individual works in the collection are in the public domain in the
United States. If an individual work is unprotected by copyright law
in the United States and you are located in the United States, we do
not claim a right to prevent you from copying, distributing,
performing, displaying or creating derivative works based on the
work as long as all references to Project Gutenberg are removed. Of
course, we hope that you will support the Project Gutenberg™
mission of promoting free access to electronic works by freely
sharing Project Gutenberg™ works in compliance with the terms of
this agreement for keeping the Project Gutenberg™ name associated
with the work. You can easily comply with the terms of this
agreement by keeping this work in the same format with its attached
full Project Gutenberg™ License when you share it without charge
with others.

1.D. The copyright laws of the place where you are located also
govern what you can do with this work. Copyright laws in most
countries are in a constant state of change. If you are outside the
United States, check the laws of your country in addition to the
terms of this agreement before downloading, copying, displaying,
performing, distributing or creating derivative works based on this
work or any other Project Gutenberg™ work. The Foundation makes
no representations concerning the copyright status of any work in
any country other than the United States.

1.E. Unless you have removed all references to Project Gutenberg:

1.E.1. The following sentence, with active links to, or other

immediate access to, the full Project Gutenberg™ License must
appear prominently whenever any copy of a Project Gutenberg™
work (any work on which the phrase “Project Gutenberg” appears,
or with which the phrase “Project Gutenberg” is associated) is
accessed, displayed, performed, viewed, copied or distributed:
 This eBook is for the use of anyone anywhere in the United
States and most other parts of the world at no cost and with
almost no restrictions whatsoever. You may copy it, give it away
or re-use it under the terms of the Project Gutenberg License
included with this eBook or online at www.gutenberg.org. If you
are not located in the United States, you will have to check the
laws of the country where you are located before using this
eBook.

1.E.2. If an individual Project Gutenberg™ electronic work is derived

from texts not protected by U.S. copyright law (does not contain a
notice indicating that it is posted with permission of the copyright
holder), the work can be copied and distributed to anyone in the
United States without paying any fees or charges. If you are
redistributing or providing access to a work with the phrase “Project
Gutenberg” associated with or appearing on the work, you must
comply either with the requirements of paragraphs 1.E.1 through
1.E.7 or obtain permission for the use of the work and the Project
Gutenberg™ trademark as set forth in paragraphs 1.E.8 or 1.E.9.

1.E.3. If an individual Project Gutenberg™ electronic work is posted

with the permission of the copyright holder, your use and distribution
must comply with both paragraphs 1.E.1 through 1.E.7 and any
additional terms imposed by the copyright holder. Additional terms
will be linked to the Project Gutenberg™ License for all works posted
with the permission of the copyright holder found at the beginning
of this work.

1.E.4. Do not unlink or detach or remove the full Project

Gutenberg™ License terms from this work, or any files containing a
part of this work or any other work associated with Project
Gutenberg™.

1.E.5. Do not copy, display, perform, distribute or redistribute this

electronic work, or any part of this electronic work, without
prominently displaying the sentence set forth in paragraph 1.E.1
with active links or immediate access to the full terms of the Project
Gutenberg™ License.

1.E.6. You may convert to and distribute this work in any binary,
compressed, marked up, nonproprietary or proprietary form,
including any word processing or hypertext form. However, if you
provide access to or distribute copies of a Project Gutenberg™ work
in a format other than “Plain Vanilla ASCII” or other format used in
the official version posted on the official Project Gutenberg™ website
(www.gutenberg.org), you must, at no additional cost, fee or
expense to the user, provide a copy, a means of exporting a copy, or
a means of obtaining a copy upon request, of the work in its original
“Plain Vanilla ASCII” or other form. Any alternate format must
include the full Project Gutenberg™ License as specified in
paragraph 1.E.1.

1.E.7. Do not charge a fee for access to, viewing, displaying,

performing, copying or distributing any Project Gutenberg™ works
unless you comply with paragraph 1.E.8 or 1.E.9.

1.E.8. You may charge a reasonable fee for copies of or providing

access to or distributing Project Gutenberg™ electronic works
provided that:

• You pay a royalty fee of 20% of the gross profits you derive
from the use of Project Gutenberg™ works calculated using the
method you already use to calculate your applicable taxes. The
fee is owed to the owner of the Project Gutenberg™ trademark,
but he has agreed to donate royalties under this paragraph to
the Project Gutenberg Literary Archive Foundation. Royalty
payments must be paid within 60 days following each date on
which you prepare (or are legally required to prepare) your
periodic tax returns. Royalty payments should be clearly marked
as such and sent to the Project Gutenberg Literary Archive
Foundation at the address specified in Section 4, “Information
 about donations to the Project Gutenberg Literary Archive
Foundation.”

• You provide a full refund of any money paid by a user who

notifies you in writing (or by e-mail) within 30 days of receipt
that s/he does not agree to the terms of the full Project
Gutenberg™ License. You must require such a user to return or
destroy all copies of the works possessed in a physical medium
and discontinue all use of and all access to other copies of
Project Gutenberg™ works.

• You provide, in accordance with paragraph 1.F.3, a full refund of

any money paid for a work or a replacement copy, if a defect in
the electronic work is discovered and reported to you within 90
days of receipt of the work.

• You comply with all other terms of this agreement for free
distribution of Project Gutenberg™ works.

1.E.9. If you wish to charge a fee or distribute a Project Gutenberg™

electronic work or group of works on different terms than are set
forth in this agreement, you must obtain permission in writing from
the Project Gutenberg Literary Archive Foundation, the manager of
the Project Gutenberg™ trademark. Contact the Foundation as set
forth in Section 3 below.

1.F.

1.F.1. Project Gutenberg volunteers and employees expend

considerable effort to identify, do copyright research on, transcribe
and proofread works not protected by U.S. copyright law in creating
the Project Gutenberg™ collection. Despite these efforts, Project
Gutenberg™ electronic works, and the medium on which they may
be stored, may contain “Defects,” such as, but not limited to,
incomplete, inaccurate or corrupt data, transcription errors, a
copyright or other intellectual property infringement, a defective or
damaged disk or other medium, a computer virus, or computer
codes that damage or cannot be read by your equipment.

1.F.2. LIMITED WARRANTY, DISCLAIMER OF DAMAGES - Except for

the “Right of Replacement or Refund” described in paragraph 1.F.3,
the Project Gutenberg Literary Archive Foundation, the owner of the
Project Gutenberg™ trademark, and any other party distributing a
Project Gutenberg™ electronic work under this agreement, disclaim
all liability to you for damages, costs and expenses, including legal
fees. YOU AGREE THAT YOU HAVE NO REMEDIES FOR
NEGLIGENCE, STRICT LIABILITY, BREACH OF WARRANTY OR
BREACH OF CONTRACT EXCEPT THOSE PROVIDED IN PARAGRAPH
1.F.3. YOU AGREE THAT THE FOUNDATION, THE TRADEMARK
OWNER, AND ANY DISTRIBUTOR UNDER THIS AGREEMENT WILL
NOT BE LIABLE TO YOU FOR ACTUAL, DIRECT, INDIRECT,
CONSEQUENTIAL, PUNITIVE OR INCIDENTAL DAMAGES EVEN IF
YOU GIVE NOTICE OF THE POSSIBILITY OF SUCH DAMAGE.

1.F.3. LIMITED RIGHT OF REPLACEMENT OR REFUND - If you

discover a defect in this electronic work within 90 days of receiving
it, you can receive a refund of the money (if any) you paid for it by
sending a written explanation to the person you received the work
from. If you received the work on a physical medium, you must
return the medium with your written explanation. The person or
entity that provided you with the defective work may elect to provide
a replacement copy in lieu of a refund. If you received the work
electronically, the person or entity providing it to you may choose to
give you a second opportunity to receive the work electronically in
lieu of a refund. If the second copy is also defective, you may
demand a refund in writing without further opportunities to fix the
problem.

1.F.4. Except for the limited right of replacement or refund set forth
in paragraph 1.F.3, this work is provided to you ‘AS-IS’, WITH NO
OTHER WARRANTIES OF ANY KIND, EXPRESS OR IMPLIED,
INCLUDING BUT NOT LIMITED TO WARRANTIES OF
MERCHANTABILITY OR FITNESS FOR ANY PURPOSE.

1.F.5. Some states do not allow disclaimers of certain implied

warranties or the exclusion or limitation of certain types of damages.
If any disclaimer or limitation set forth in this agreement violates the
law of the state applicable to this agreement, the agreement shall be
interpreted to make the maximum disclaimer or limitation permitted
by the applicable state law. The invalidity or unenforceability of any
provision of this agreement shall not void the remaining provisions.

1.F.6. INDEMNITY - You agree to indemnify and hold the Foundation,

the trademark owner, any agent or employee of the Foundation,
anyone providing copies of Project Gutenberg™ electronic works in
accordance with this agreement, and any volunteers associated with
the production, promotion and distribution of Project Gutenberg™
electronic works, harmless from all liability, costs and expenses,
including legal fees, that arise directly or indirectly from any of the
following which you do or cause to occur: (a) distribution of this or
any Project Gutenberg™ work, (b) alteration, modification, or
additions or deletions to any Project Gutenberg™ work, and (c) any
Defect you cause.

Section 2. Information about the Mission

of Project Gutenberg™
Project Gutenberg™ is synonymous with the free distribution of
electronic works in formats readable by the widest variety of
computers including obsolete, old, middle-aged and new computers.
It exists because of the efforts of hundreds of volunteers and
donations from people in all walks of life.

Volunteers and financial support to provide volunteers with the

assistance they need are critical to reaching Project Gutenberg™’s
goals and ensuring that the Project Gutenberg™ collection will
remain freely available for generations to come. In 2001, the Project
Gutenberg Literary Archive Foundation was created to provide a
secure and permanent future for Project Gutenberg™ and future
generations. To learn more about the Project Gutenberg Literary
Archive Foundation and how your efforts and donations can help,
see Sections 3 and 4 and the Foundation information page at
www.gutenberg.org.

Section 3. Information about the Project

Gutenberg Literary Archive Foundation
The Project Gutenberg Literary Archive Foundation is a non-profit
501(c)(3) educational corporation organized under the laws of the
state of Mississippi and granted tax exempt status by the Internal
Revenue Service. The Foundation’s EIN or federal tax identification
number is 64-6221541. Contributions to the Project Gutenberg
Literary Archive Foundation are tax deductible to the full extent
permitted by U.S. federal laws and your state’s laws.

The Foundation’s business office is located at 809 North 1500 West,

Salt Lake City, UT 84116, (801) 596-1887. Email contact links and up
to date contact information can be found at the Foundation’s website
and official page at www.gutenberg.org/contact

Section 4. Information about Donations to

the Project Gutenberg Literary Archive
Foundation
Project Gutenberg™ depends upon and cannot survive without
widespread public support and donations to carry out its mission of
increasing the number of public domain and licensed works that can
be freely distributed in machine-readable form accessible by the
widest array of equipment including outdated equipment. Many
small donations ($1 to $5,000) are particularly important to
maintaining tax exempt status with the IRS.

The Foundation is committed to complying with the laws regulating

charities and charitable donations in all 50 states of the United
States. Compliance requirements are not uniform and it takes a
considerable effort, much paperwork and many fees to meet and
keep up with these requirements. We do not solicit donations in
locations where we have not received written confirmation of
compliance. To SEND DONATIONS or determine the status of
compliance for any particular state visit www.gutenberg.org/donate.

While we cannot and do not solicit contributions from states where

we have not met the solicitation requirements, we know of no
prohibition against accepting unsolicited donations from donors in
such states who approach us with offers to donate.

International donations are gratefully accepted, but we cannot make

any statements concerning tax treatment of donations received from
outside the United States. U.S. laws alone swamp our small staff.

Please check the Project Gutenberg web pages for current donation
methods and addresses. Donations are accepted in a number of
other ways including checks, online payments and credit card
donations. To donate, please visit: www.gutenberg.org/donate.

Section 5. General Information About

Project Gutenberg™ electronic works
Professor Michael S. Hart was the originator of the Project
Gutenberg™ concept of a library of electronic works that could be
freely shared with anyone. For forty years, he produced and
distributed Project Gutenberg™ eBooks with only a loose network of
volunteer support.
Project Gutenberg™ eBooks are often created from several printed
editions, all of which are confirmed as not protected by copyright in
the U.S. unless a copyright notice is included. Thus, we do not
necessarily keep eBooks in compliance with any particular paper
edition.

Most people start at our website which has the main PG search
facility: www.gutenberg.org.

This website includes information about Project Gutenberg™,

including how to make donations to the Project Gutenberg Literary
Archive Foundation, how to help produce our new eBooks, and how
to subscribe to our email newsletter to hear about new eBooks.
Welcome to our website – the perfect destination for book lovers and
knowledge seekers. We believe that every book holds a new world,
offering opportunities for learning, discovery, and personal growth.
That’s why we are dedicated to bringing you a diverse collection of
books, ranging from classic literature and specialized publications to
self-development guides and children's books.

More than just a book-buying platform, we strive to be a bridge

connecting you with timeless cultural and intellectual values. With an
elegant, user-friendly interface and a smart search system, you can
quickly find the books that best suit your interests. Additionally,
our special promotions and home delivery services help you save time
and fully enjoy the joy of reading.

Join us on a journey of knowledge exploration, passion nurturing, and

personal growth every day!

ebookmasss.com