planning in intelligent systems

The following five factors will be recurring themes throughout this book: ning entity, model of the future, planning process, executing entity, and plan.. The first is a model of the decis

PLANNING IN

INTELLIGENT SYSTEMS Aspects, Motivations, and Methods

PLANNING IN

INTELLIGENT SYSTEMS

PLANNING IN

INTELLIGENT SYSTEMS Aspects, Motivations, and Methods

Copyright # 2006 by John Wiley & Sons, Inc All rights reserved.

Published by John Wiley & Sons, Inc., Hoboken, New Jersey.

Published simultaneously in Canada.

No part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or

by any means, electronic, mechanical, photocopying, recording, scanning, or otherwise, except as permitted under Section 107 or 108 of the 1976 United States Copyright Act, without either the prior written permission of the Publisher, or authorization through payment of the appropriate per-copy fee

to the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, (978) 750-8400, fax (978) 750-4470, or on the web at www.copyright.com Requests to the Publisher for permission should be addressed to the Permissions Department, John Wiley & Sons, Inc., 111 River Street, Hoboken, NJ 07030, (201) 748-6011, fax (201) 748-6008, or online at http://www.wiley.com/go/

Limit of Liability /Disclaimer of Warranty: While the publisher and author have used their best efforts in preparing this book, they make no representations or warranties with respect to the accuracy or completeness of the contents of this book and specifically disclaim any implied warranties of merchantability or fitness for a particular purpose No warranty may be created or extended by sales representatives or written sales materials The advice and strategies contained herein may not be suitable for your situation You should consult with a professional where appropriate Neither the publisher nor author shall be liable for any loss of profit or any other commercial damages, including but not limited to special, incidental, consequential, or other damages.

For general information on our other products and services or for technical support, please contact our Customer Care Department within the United States at (800) 762-2974, outside the United States at (317) 572-3993 or fax (317) 572-4002.

Wiley also publishes its books in a variety of electronic formats Some content that appears in print may not be available in electronic formats For more information about Wiley products, visit our web site at www.wiley.com.

Library of Congress Cataloging-in-Publication Data:

Planning in intelligent systems: aspects, motivations, and methods /edited

by Wout van Wezel, Rene Jorna, Alexander Meystel.

p.cm

Includes bibliographical references and index.

ISBN 0-471-73427-6 (cloth)

1 Expert systems (Computer science) 2 Intelligent control systems 3.

Artificial intelligence I Wezel, Wout van II Jorna, Rene III Meystel,

A (Alex)

QA76.76.E95P533 2006

006.303- -dc22

2005021353 Printed in the United States of America

10 9 8 7 6 5 4 3 2 1

permission.

2 How We Do What We Want: A Neurocognitive Perspective on

4 Cognition, Planning, and Domains: An Empirical Study into

Rene´ Jorna

v

Wout van Wezel

Leo G Kroon and Rob A Zuidwijk

Ju¨rgen Sauer

Michael H Bowling, Rune M Jensen, and Manuela M Veloso

11 Multiresolutional Representation and Behavior Generation:

How Does It Affect the Performance of and Planning for

Alexander Meystel

12 Perspectives on Shunting Planning: Research in Planning

Wout van Wezel and Derk Jan Kiewiet

13 Task Analysis for Problems of Shunting Planning within the

Derk Jan Kiewiet, Rene´ Jorna, and Wout van Wezel

14 Intelligent Shunting: Dealing with Constraints (Satisfaction) 391Erwin Abbink

15 Applying Operations Research Techniques to Planning of

Ramon M Lentink, Pieter-Jan Fioole,

Leo G Kroon, and Cor van’t Woudt

16 Train Shunting: A Practical Heuristic Inspired by

R Haijema, C.W Duin, and N.M van Dijk

17 Planner-Oriented Design of Algorithms for Train

J Riezebos and Wout van Wezel

18 Conclusions for Intelligent Planning: Diversity and the

Rene´ Jorna, Wout van Wezel, and Alexander Meystel

R Hajema, Faculty of Economics and Econometrics, Universiteit van Amsterdam,P.O Box 19268, 100066, Amsterdam, the Netherlands

Jean-Michel Hoc, Centre National de la Recherche Scientifique et Universite´ deNantes, F-44321 Nantes, France

Bernhard Hommel, Cognitive Psychology Unit, Department of Psychology,Leiden University, 2300-RB Leiden, the Netherlands

Rune M Jensen, Computer Science Department, Carnegie Mellon University,Pittsburgh, PA 15213 – 3891

ix

Rene´ Jorna, Faculty of Management and Organization, University of Groningen,NL-9700-AV Groningen, the Netherlands

Derk Jan Kiewiet, Faculty of Management and Organization, University ofGroningen, NL-9700-AV Groningen, the Netherlands

Leo G Kroon, Rotterdam School of Management (RSM), Erasmus UniversityRotterdam, NL-3000-DR Rotterdam, the Netherlands; and Department ofLogistics, NS Reizigers, NL-3500-HA Utrecht, the Netherlands

Ramon M Lentink, Rotterdam School of Management (RSM), ErasmusUniversity Rotterdam, NL-3000-DR Rotterdam, the Netherlands

Kenneth N McKay, Department of Management Sciences, University ofWaterloo, Waterloo, Ontario, Canada N2L 3G1

Alexander Meystel, Electrical and Computer Engineering Department, DrexelUniversity, Philadelphia, PA 19104

J Riezebos, Faculty of Management and Organization, University of Groningen,NL-9700-AV

Ju¨rgen Sauer, Department of Computer Science, University of Oldenburg,D-26121 Oldenburg, Germany

N.M van Dijk, Faculty of Economics and Econometrics, Universiteit vanAmsterdam, P.O Box 19268, 100066, Amsterdam, the Netherlands

Wout van Wezel, Faculty of Management and Organization, University ofGroningen, NL-9700-AV Groningen, the Netherlands

Cor van’t Woudt, Department of Logistics, NS Reizigers, NL-3500-HA Utrecht,the Netherlands

Manuela M Veloso, Computer Science Department, Carnegie Mellon University,Pittsburgh, PA 15213 – 3891

Vincent C.S Wiers, Institute for Business Engineering and Technology cation (BETA), Eindhoven University of Technology, Eindhoven 5600-MB,the Netherlands

Appli-Rob A Zuidwijk, Rotterdam School of Management (RSM), Erasmus UniversityRotterdam, NL-3000-DR Rotterdam, the Netherlands

To be able to plan, one needs intelligence To act intelligently, however, one needs

to plan The questions that this paradox raises express the goal of this book Theabilities to anticipate and plan are essential features of intelligent system, whetherthey are human or machine We might even go further and contemplate that abetter planning results in higher achievements As a consequence, understandingand improving planning is important So, how do intelligent systems make plans?What are their motivations? How are the plans executed? What is the relationbetween plan creation and plan execution? Are planning and intelligence as con-nected as the paradox suggests? Many questions that are studied by a manifold

of research disciplines for various kinds of intelligent systems employ a wealth ofplanning goals, methods, and techniques Our goal is to investigate whether planningapproaches in the different fields share more than merely the word planning If so,knowing of the various planning paradigms might lead to a better understanding

of planning in general and to cross-fertilization of ideas, methods, and techniquesbetween the different planning schools in, for example, cognitive psychology,organizational science, operations research, computer science, and robotics

In September 1999, we (Wout, Rene´, and Alexander) organized a session onplanning at the International Congress of the IASS-AIS (International Associationfor Semiotic Studies) in Dresden At that meeting, we discussed the lack of a generalplanning theory, and we decided that a book discussing and comparing the various(scientific) fields that deal with planning would be a good start The ideas about ageography, a landscape, of planning were further explored at a session at theINFORMS (Operations Research) meeting in Philadelphia in November 1999.There, we invited a number of people with the prospect to further discuss andexplore the ideas about the book

xi

The conferences at which we discussed the book and the people involved inwriting the chapters are emblematic for the broad scope of planning approachesthat exist Our own backgrounds resemble this as well The throughput time ofsix years might very well be ascribed to the different disciplines in which our indi-vidual backgrounds can be found Discussions between a control theorist and expert

in cybernetics, a cognitive scientist, and a production management and tional scholar that should lead to a single comprehensive book on planning areprone to lead to intense discussions about the approaches in particular We are con-vinced that broadening the scope as we did is necessary as a first step in finding aunified theory of planning

organiza-We are indebted to several people that helped us in various ways in the creation

of the book The Netherlands Railways—in particular, Tjeu Smeets and LeoKroon—provided ample opportunities for valuable empirical research in a stimulat-ing and challenging environment Jan Riezebos and Herman Balsters reviewedsome of the mathematical chapters Sonja Abels provided operational support andRene´ Jorna thanks NIAS (Netherlands Institute for Advanced Studies; KNAW),where he stayed the academic year of 2004/2005, for providing support andfacilities

RENE´ JORNA

ALEXANDER MEYSTEL

INTRODUCTION

WOUT VAN WEZEL AND RENE´ JORNA

Faculty of Management and Organization, University of Groningen,

NL-9700-AV Groningen, the Netherlands

1.1 INTRODUCTION

No living thing seems to be conscious of the future, and none seems concerned todesign for that future, except Man But every man looks ahead and attempts to organizefor tomorrow, the future of the next day or of the next generation Whatever he has to

do or proposes to do, he plans; he is a planner He seems to be distinguished from allother forms of life by this faculty, this necessity Man plans to rebuild employment, or

to increase his company’s volume of business, or to win an election, or to write a letter,

or to build a bridge, or to buy a cigar, or to get his hair cut, or to put alcohol in the tor of his car against the prospect of sub-zero weather, or to give the baby paregoricagainst the prospect of a sleepless night, or to build a city—he plans all the time Byhis very nature every man plans constantly

radia-—Jacob L Crane, Planning organization and the planners, paper presented atthe Annual Meeting of the American City Planning Institute, WashingtonD.C., January 19, 1936

In his paper, Crane provides us with an analysis of the differences between ual personal planning and city planning by governmental agencies He concludesthere are similarities, but there are sharp differences as well For example, ahuman planning for himself both determines the course of action, and he acts him-self In contrast, a city planner in a government will only make the plan, not execute

individ-it Furthermore, coordination and integration of plans is more important for cityplanning than for personal planning In the decades following Crane’s observations,

much research has been done on planning We now know of the physical andcognitive functions that humans use to make plans for themselves Furthermore,planning is a formalized function in almost all organizations, and much literaturehas been written about how plans in organizations can or should be made.

An important event for planning that could probably not be foreseen in the ning of the twentieth century, when Crane wrote about planning, is the widespreaduse of computers Computer programs can make plans as well as individuals andorganizations Examples are algorithms in advanced planning systems that createschedules for all kinds of organizational processes (e.g., production schedules,staff schedules, routing schedules), and planning algorithms that are used byrobots to play soccer or to collect stones on Mars

begin-Research in planning has mainly been categorized by taking both the creator ofthe plan and who it is created for as a starting point For example, how a humanmakes a shopping list is investigated in academia other than that for a humanwho makes a production schedule or an unmanned air vehicle that determines itsown route In this book we explore various planning approaches and we try to deter-mine whether planning can be a self-sufficient area of research where scholars fromdifferent disciplines can exchange ideas and share research results Sharing researchmethodologies, planning methods, and solution techniques will improve our under-standing of planning and scheduling in general, and it can result in improvements ineach of the individual planning research schools

In this introductory chapter, we will present our conviction that different planningfields share many characteristics Each of the subsequent chapters will describe anddiscuss a specific kind of planning In the concluding chapter, we will formulate thesimilarities and, of course, the differences and we will show the prospects of acommon research agenda

In this chapter we start in Sections 1.2, 1.3, and 1.4 with discussions about whatplanning is and how it can be modeled Sections 1.5 and 1.6 describe a number ofcharacteristics of planning These characteristics can be used as a first startingpoint to compare planning approaches Section 1.7 outlines the structure of the book

1.2 DEFINITION OF PLANNING

Where will we go and how do we get there? This question is an inherent part of thefunctioning of humans and organizations The ability to anticipate and plan isusually seen as a required and perhaps even essential feature of intelligent systems

It is the fundament of goal-directed behavior; systems that pursue goals need to takethe future into account In this book, we will compare different planning researchfields To be able to compare and analyze differences and similarities, we need acommon, abstract conceptualization of planning As a starting point, we presumethat all intelligent systems use anticipation to plan (van Wezel and Jorna, 2001)

An anticipatory system is “a system containing a predictive model of itself and/

or its environment, which allows it to change a state at an instant in accordance withthe model’s predictions pertaining to a later instant” (Rosen, 1985) Our definition

of planning will be built around this definition of anticipatory systems, bydistinguishing three main elements of planning.

First, it is important to acknowledge that some entity must make the plan Notethat all kinds of entities—for example, humans, robots, computer programs, ani-mals, organizations, and so on—can make plans Important features of the planningentity are (a) the model of the future that the planning entity has and (b) the processcharacteristics of making the plan:

a The planning entity needs some kind of model of the future, since the future isessentially nonexistent This model should include states, possible actions ofthe executing entities and the effect of actions on the state they reside in, con-straints, and goals Planning and anticipation presume that such a predictivemodel is available; otherwise, the chance that a plan can be executed asintended becomes a shot in the dark

b Planning is a process It consists of all kinds of activities that ultimately result

in the plan Information must be collected, there might be communicationabout constraints, difficult puzzles must be solved, and so on The kinds ofprocesses are determined by the kind of entity that makes the plan, butthere are many generic characteristics as well

Second, someone or something must execute the plan; that is, the intended futuremust somehow be attained Again, this can be done by all kinds of entities, and theplanning entities need not necessarily be involved in plan execution themselves.The third element of planning is the plan itself The plan is the main communi-cation mechanism between the planning entity and planned entity The plan signifiesthe belief that the planning entity has in the model of the future: The implicit or expli-cit actions in the plan will lead to the desired or intended future state It can never be afull specification of the future itself because it can never be specified more preciselythan the model of the future allows It can, of course, be specified with less detail thanthe model of the future Two kinds of plans are possible First, the plan can specifythe intended future state The executing entity itself must determine how to get there.Second, the plan can specify the actions that the executing entity must perform.Although the desired future state is then not specified in the plan as such, it will,ceteris paribus, be reached by performing all specified actions

The following five factors will be recurring themes throughout this book: ning entity, model of the future, planning process, executing entity, and plan Wewill see that this definition provides a sound basis with which planning approachescan be described and compared The following sections will describe some moregeneric aspects of these elements

plan-1.3 PLANNING COMPLEXITY AND PLANNING HIERARCHIES

A widely accepted characteristic of planning is that it is complex This book wouldnot be necessary if planning were trivial and easy to understand But then, what is so

complex about planning? Humans plan their errands continuously, production ners schedule whole factories, automatic vehicles find their destination, and evenmicroprocessors plan the execution of computer code to increase speed Apparently,something strange is going on On the one hand, plans are made all the time On theother hand, we find it difficult to understand the way that humans make plans and todesign systems or computer programs that plan The partial answer to this, of course,

is that planning problems do not exist by themselves, but are perceived by the ning entity This entity, which we presume intelligent, will not formulate unsolvableproblems for itself, or, as Simon (1981, p 36) states: “What a person cannot do hewill not do, no matter how much he wants to do it.” In addition, the model of thefuture is full of uncertainties, and even for a problem that the planning entityitself has formulated, good-enough alternatives will be accepted, “not because heprefers less to more but because he has no choice” (op cit.) This is inevitablythe case for planning, because time is not only a part of the plan, it is also somethingthat is used up during plan creation The moment of plan execution is getting nearerand nearer while the planning entity is seeking the solution At some point in time,whether the planning entity is happy with it or not, the plan must be executed Some-how, intelligent systems know how to make planning simple enough to be manage-able but complex enough to attain advantageous goals

plan-Simon (1981) notes that complex systems are usually somehow ordered archically in order to manage complexity He uses the term hierarchy in the sensethat a system is “composed of interrelated subsystems, each of the latter being inturn hierarchic in structure until we reach some lowest level of elementary subsys-tems” (op cit., p 196) Note that this does not necessarily mean hierarchic in thesense of an authority relation; it means an ordering of parts in wholes, and thesewholes are, in turn, parts of other wholes We will argue that planning is no excep-tion; setting aside trivial planning problems, planning always takes place hierarchi-cally Even more, we will argue that much of the differences between planningapproaches can be contributed to the way in which these approaches partition theplanning problem in independently solvable subproblems Therefore, a sound under-standing of the hierarchical nature of planning is a prerequisite for understanding thedifferences and similarities between planning approaches

hier-There are two main reasons to take planning decisions in a hierarchy(Starr, 1979) First, some decisions must be made hierarchically due to a lack ofinformation This means that a decision is needed before all the required input forthat decision is available The input must then somehow be predicted An example

is that a company must order raw materials before their own customers place theirorders They do this on the basis of an expectation about the total amount of orders,which is a different hierarchical level than the individual order Second, decisionscan be taken hierarchically because it reduces the amount of information that a plan-ning entity must process Planning problems are often transcomputational, whichmeans that the amount of plan alternatives is so large that even a computer that isthe size of the earth cannot assess all possibilities in millions of years (Klir,1991) As an example, consider the simple task of determining the sequence of 20tasks If the planning entity (for example, a computer program) can assess 1 billion

sequences (plans) per second, it will take 77 years to check all possible sequencesand will take 1620 years with 21 operations For that reason, most planning prob-lems cannot be solved by assessing all possible plans and choosing the best Inorder to limit the work, a plan can be created in a hierarchical way.

A plan consists of statements about the future As we saw, a plan can eitherspecify a future state description or actions that lead to that state Notwithstandingthe manifestation of the plan, the creation of it always involves decision-making If

we view the process of plan creation as a system, every decision that somehow mines a part of this future can be regarded as a subsystem An example of such ahierarchy of planning is the creation of a plan that assigns orders to machines Ifmaking the total plan is seen as a system, then the assignment of an order to amachine for Tuesday between 12AMand 3PMcan be seen as a subsystem (Meysteland Albus, 2002) This assignment itself can also be seen as a system, which is com-posed of subsystems or subdecisions With the use of this paradigm, we can usesystem theories for analyses of planning decisions by intelligent systems Theview of planning as a hierarchy of decisions provides a common ground for all plan-ning decisions that are made regardless of the level of detail

deter-A consequence of the view that a hierarchy exists in decision making is that ahierarchy also exists in the things that are planned More specifically, the modelthat the planning entity has of the planned entity and its environment must allowhierarchical decision making This implies that the model of the future must alsoallow descriptions at hierarchical levels In the next section, we will elaborate onthis by discussing generic models of decision behavior and the planning domain

1.4 BASIC MODELS OF PLANNING

There are two kinds of basic models for all planning situations The first is a model

of the decision behavior of the planning entity, and the second is a model of theresults of the decision behavior (the plan) We will discuss both in greater detail

1.4.1 Making the Plan: Decisions of the Planning Entity

In Section 1.2 we have discussed that planning involves a planning entity In Section1.3, we specified that such a planning entity makes decisions, and that thesedecisions are ordered hierarchically In this section, we propose a generic decisionmodel that is based on these principles A consequence of the generic nature of thismodel is that it does not describe in detail how planning decisions are related to otherkinds of decisions In Chapter 4, Jorna goes into this issue by analyzing and discuss-ing the differences between planning, decision making, and problem solving

As stated in the previous section, we view the planning process as a hierarchy ofdecision-making activities At each hierarchical level, the task is to find a solutionwithin the constraints that are specified by the higher level A constraint is a rulethat restricts the possible plan alternatives Constraints can be determined before-hand as inherent parts of the model of the future, but they can also be determined

during the process of plan creation Often, the higher level also specifies the goalsthat should be attained as good as possible There are two kinds of constraintsand goals:

1 Content Constraints and Goals These constraints and goals relate to the ution itself They can specify aggregates (e.g., the average number of hoursthat employees can work, the minimum utilization rate of the machines, theamount of food I must buy for next week at the grocery, etc.) or specificrules (e.g., John and Jack may not be in the same shift, I must buy exactlyone loaf of bread, etc.)

sol-2 Process Constraints and Goals These restrain the way in which the planningentity makes the plan This can be about the maximum throughput time formaking the plan, about the maximum amount of information processingcapacity that may be used (for example, the number of planners that isinvolved), about the tools that can be used, whether a factory planner maynegotiate directly with customers about due dates or not, and so on

The solution at a given hierarchical level specifies the constraints and goals forthe lower level(s) In this way, using multiple hierarchical levels, the plan can bemade stepwise An important feature of a hierarchical decision-making system isthat decision levels should be able to handle feedback If a constraint is toosevere and the lower level cannot find a solution, constraints must be relaxed.The decisions at the lowest level are not specified in greater detail by the planningentity The solution at the lowest level specifies when, what, and how the plannedactions must be executed However, most often the plan must be specified furtherduring the execution of the plan For example, a plan can specify that the production

of a batch in a factory starts at 4PM Usually, this does not mean that it will start

at exactly 4PM In factory settings, it usually will start when the batch that wasscheduled as its immediate predecessor (e.g., the one starting at 3PM) is finished.Figure 1.1 depicts the discussed elements It shows three decisions with theirrelations, and reveals the following characteristics:

. A hierarchical planning decision is defined as a decision that constrains anotherdecision (arrow 1) Therefore, the hierarchical relation between two decisions

is based on the fact that a decision’s solution space is restricted by the otherdecision

. It might be difficult or impossible to make a decision within the imposedrestrictions Then, somehow this must be fed back to the decision that imposedthe constraint (arrow 2)

. Decisions that share constraints must somehow be coordinated because theircombined decisions determine whether the constraint is violated or not(arrows 3 and 4)

Figure 1.2 shows an example of a decision structure that is based on this model

Figure 1.3 shows a decision structure at the individual level for the shunting blem in the Netherlands Railways An individual planner has to go through a (struc-tured) hierarchy of decisions to produce a plan for trains at the shunting yard.

pro-A collection of decisions with their hierarchical relations constitutes the way that

a planning problem is tackled The model in Figure 1.1 shows the basic elementsthat can be used to model the structure of planning levels This view on planningimplies that relations between planning decisions are always structurally the sameregardless of the decision level and regardless of the entities that make and executethe plan In Section 1.5, we will describe in greater detail more general character-istics of planning decisions First, however, we will discuss what it is that planningdecisions decide about

1.4.2 Modeling the Plan: States of the Planning Entity

As stated in the previous subsection, planning decisions differ from other kinds ofdecisions We can now describe (at least partly) what planning decisions are bydescribing the decision domain First, planning is a synthetic rather than an analytic(diagnosis) or modification (repair) task (Clancy, 1985; Schreiber, Wielinga, andBreuker, 1993) Second, planning involves decisions about the future and not theexecution of these decisions Third, an important feature of planning is that it isabout choosing one alternative out of a huge number of alternatives that are struc-turally similar Determining why a motor does not work is not planning (it is a diag-nosis task), building a house is not planning (it is also a synthetic task: however, itconcerns not only a decision, but also the realization of the plan), but routing trains,making a production schedule, making a staff schedule, and determining the trajec-tory of an automatic vehicle are planning tasks (these are synthetic tasks and concernchoosing one out of a number of similar alternatives of future states)

Decision

1 2

1 2

Figure 1.1 Basic hierarchic decision model

With this demarcation, we can further define planning, by explaining what ismeant by “structurally similar alternatives.” The vague connotation of the word

“similar” already indicates that it is not inherently clear whether a problem is a ning problem or not, but that in itself is not important We propose to model a plan-ning problem as follows A planning problem consists of groups of entities, wherebythe entities from different groups must be assigned to each other The assignmentsare subject to constraints, and alternatives can be compared on their level ofgoal realization For example, production scheduling is a problem where ordersmust be assigned to machines, in a shift schedule people are assigned to shifts,and in task planning tasks are assigned to time slots and resources Now we can

plan-Aggregate balancing

of suppliers, customers, and

capacity

Aggregate balancing

of products and capacity

Long horizon product

planning

Order management

Week pattern determination

Other track same interval?

Earlier or later through area?

Other track other interval?

Depart from other track?

Skip internal washing

Skip external washing

Do the driver and/or shunter

have time for the altered job?

8 10

True

Apply the solution and

go to next event

Shunting event x

Select a train to free the driver’s and/or shunter’s time False

Figure 1.3 Example of a decision structure (individual level)

also specify what we mean by “similar”; it means that plan alternatives have the samestructure (e.g., orders are assigned to machines), but a different content (e.g., in planalternative A, “order 1” is assigned to “machine 1,” and in plan alternative B, “order1” is assigned to “machine 2”) This definition also precludes some areas that arecommonly regarded as planning—for example, strategic planning and retirementplanning Although the boundaries are debatable, such planning problems do notexhibit the third feature of planning—that is, alternatives are structurally similar.Not by coincidence, the view on planning as just described fits in nicely with thedecision models, where basic decisions are setting constraints and assigning entities.The link to decision models, however, shows an additional requirement If decisions

at hierarchical levels are distinctive, then there should also be models of domains atmultiple hierarchical levels For example, in Figure 1.2, each decision deals witheither (a) other kinds of planned entities or (b) planned entities at different levels

of abstraction or resolution Furthermore, the planning entity can be an aggregate

of planning entities itself Most apparently, this is the case in organizational ning, where a planning department can be said to make a plan, but where individualhuman planners make the subplans, as is the case in Figure 1.2

plan-Two types of subplans can be distinguished in a planning hierarchy: aggregationand decomposition In aggregation, the dimensions that exist in the planning problemstay the same, but individual entities of a dimension are grouped For example, a planthat contains the assignment of individual orders to production lines for a certainweek can be aggregated to a plan that contains the assignment of orders per producttype to production lines in that week Aggregation can be used to establish boundaries

or constraints for individual assignments of entities that fall within an aggregatedgroup For example, it is first decided how much caramel custard will be madenext week Then individual orders that fall in this product family can be assigned

to a specific production time In this way, several stages of aggregation can besequentially followed whereby each stage creates boundaries for the next stage

In the second type of subplan, decomposition, a subset of the entities that must beplanned is considered as a separate planning problem Decomposition can deal withall entities of a subset of the dimensions, all dimensions with a subset of the entities,

or a combination of subsets of dimensions and entities For example, if we attuneorders, machines, and operators, we could first assign orders to machines and thenoperators to the chosen order/machine combinations Or, we can first assign allcustomer orders, after which we assign all stock orders

The decision models and the state models of planning together can be used todepict the decision behavior of planning entities In the next subsection, we willrecapitulate the elements that form our basic model of planning

1.4.3 Conclusion

In Sections 1.2 and 1.3, we have described what planning is:

. Planning means that a planning entity determines a future course of actions for

an executing entity These actions should lead to a desired future state The

belief that the actions lead to the state is based on the model of the future of theplanning entity The future course of actions or the desired future state isexpressed by the plan.

. Planning is a complex activity and often involves reasoning with incompleteinformation Plans are usually made hierarchically

Then, in this Section (1.4), we have further explored generic features of planning, byproposing how decisions of the planning entity can be modeled:

. A plan contains the assignments of entities of different categories

. The assignments are subject to constraints

. Alternatives can be compared on their level of goal realization

. During the process of plan creation, subplans can be created at hierarchicallevels other than the final plan

. Constraints and goals are distinct at each hierarchical level

. Decisions determine constraints for lower-level decisions

. Grouping takes place by aggregation

. Partitioning takes place by disaggregation or by decomposition

Figure 1.4 summarizes the elements of planning

In this book a number of planning approaches are discussed by various authors(Table 1.1) Although the approaches at first sight do not always seem to havemuch in common, our generic planning model provides the means to compare theapproaches and analyze where they differ Thus, we can show that differentapproaches share more than they differ

The aim of this book is to show that planning approaches can be categorizedalong other dimensions than their research field In the next section, we will describe

a number of generic characteristics that can be derived from this model Theapproaches and their relation to these generic characteristics will be discussed inmore detail in the individual chapters of the book The diversity in the planningapproaches will become clear by stating questions that are based on the model inFigure 1.4, for example:

1 What is the planning entity? Is it a natural entity (i.e., human) or an artificialentity? How does it make decisions? How is the planning decomposed? Whatare the partitioning criteria? In what order are the decisions made?

2 What is the executing entity? Is it an organization or an individual? Is itperhaps the planning entity itself ? Do multiple executing entities have tocoordinate or are they independent?

3 What kind of model of the future does the planning entity have? How flexible

is the model with respect to adjustment?

The following sections describe a number of generic characteristics of planning ties that can be used to formulate answers to these kinds of questions.

enti-1.5 GENERIC PLANNING CHARACTERISTICS

In this section, we describe a number of information processing characteristics ofthe planning entity and its relation to the environment The characteristics thatwill be discussed are: (a) closed versus open world assumptions; (b) the informationprocessing mechanism and its architectural components, such as memory or atten-tion; (c) (internal) representations; (d) communication, meaning, and interpretation;(e) characteristics of coordination; and (f ) aspects of execution of the plan

“Closed World” Versus “Open World” As we already indicated, the planning taskitself can be called a synthetic or configuration task It is well known that these kinds

Planning entity

Subplan (constraints and goals)

feedback

Figure 1.4 Overview of planning

of tasks are very difficult to complete, by humans alone as well as with the support ofsoftware In the previous section, we described a way to model plans and planningdecisions Each assignment problem consists of choosing from alternatives that arestructurally the same In classical terms this means searching through a problemspace to find an acceptable solution, which is usually called problem solving.From a decision perspective, realizing a suitable plan or solving a planning problemrequires three nearly decomposable phases In state space descriptions the first phase

is the design of a complex initial state, of goal states, and of admissible operations tochange states Note that we are talking here about the states in the search process of

TABLE 1.1 Planning Approaches Discussed in This Book

of human planning

Catching a ball

aspects of human planning

Flying a fighter aircraft

aspects of humans planningfor organizations

Making a staff schedule

multi-actor systems

Two people that give adifferent meaning to aplan

McKay and Wiers (Chapter 6) Planning processes in

manufacturingorganizations

The relations betweenplanning and functionaldepartments likepurchasing, sales, andproduction

organizations

The trade-off betweengoal maximization andusability of analgorithmKroon and Zuidwijk

(Chapter 8)

Mathematical algorithms forplanning support inorganizations

An algorithm to generate atime table for a railwayoperator

algorithms for planningsupport in organizations

Algorithms that makeschedules thatcoordinate activities inmultiple sites of afactory

Bowling, Jensen, and

Veloso (Chapter 10)

Artificial Intelligencealgorithms for planning byrobots

Two robots that playsoccer and create acoordinated plan

multiresolutional controlhierarchy

The role of planning in thereasoning process of anintelligent actor

the planning entity and not about the states that the planned domain (or the plannedentity) can be in The second phase is, given the admissible operations, to search for

an optimal solution The search process may be done by exhaustive computation or

by adequate heuristic evaluation functions in combination with sophisticated putations In many cases, search does not give an optimal solution The most onemay get is a satisfying solution, and even that is often not possible Then, thethird phase starts in which one goes back to the initial state and the admissible oper-ations Another route is chosen in the hope that a solution is found In other words,the phases of (1) initial state, (2) search, no solution, and (3a) start again with initialstate follow the so-called “closed world” assumption This is the necessary sequence

com-if algorithms are applied However, there is another way of dealing with the thirdphase which is more usual, especially if humans have to make a planning If,given the constraints and goal functions, the second phase does not give an optimal

or satisfactory outcome, the planner is already so much involved in the planning cess that because he has a glimpse of the solution given the constraints, he takes his

pro-“idea” of a solution for compelling He therefore changes the initial state(s) andadmissible operations—that is, the constraints—in such a way that they fit the pre-conceived solution This order of phases can be named the “open world” approach Itconsists of (1) initial state, (2) search, including not finding a real or establishedfixed solution, and (3b) adjustment of initial state according to the “fixed” solutionreality In other words, the model of the future is not fixed because rules are adjusted.This sequence of activities is what human planners—whether in the industry, in trans-portation planning, or in staff scheduling—frequently and with great success do.However, formalizing such knowledge for use in a computer program or robot isvery tricky In AI this kind of practice is also known as the reformulation problem

It remains an open issue how and from which perspective the planning entity adjustshis initial state and/or operations

Information Processing Mechanism During problem solving, the planning entityhas to process information An information processing mechanism operationalizesthe way information is selected, combined, created, and deleted The mechanismitself needs a physical or physiological carrier Examples are the brain as our neuro-logical tissue, the layered connection system of a chip in a computer, a human indi-vidual in an organization, or a group of interconnected individuals in anorganization This is of course relevant when we realize that the contents of themodel of the future can be restricted by the physical, physiological, or functionalproperties of the carrier

The most relevant distinction is the one in internal and external mechanism Byinternal we mean that there is no direct access to the system from outside Internallycontrolled, but not directly visible, processes—not cognitively penetrable asPhylyshyn (1984) called it—take place in the system The cognitive system andthe chip are internal, but they differ in the sense that the latter is designed, whichmeans that its operations are verifiable External are information processing mech-anisms such as groups of individuals or organizations With respect to planning,this distinction is of course relevant if one realizes that if the plan needs to be

communicated, a translation is necessary between the physical carrier and the ver, which must be taken into account during planning This is the case when a plan-ning entity makes a plan that is executed by (many) others.

recei-Architectural Components An architecture is a set of components of which thearrangement is governed by principles of form or function (Curry and Tate,1991) A cognitive architecture consists of memory components, of attention pro-cessors, of sensory and motor components, and of various kinds of central pro-cessors The division is by function, and the components are all implemented inneurological structures in the brain Two other material structures for architecturallayout are the chip and the constellation of a group of individuals The same kind

of components can be discerned for the computer (e.g., a robot), consisting ofmemory, sensory and motor components, and central processors For a group of indi-viduals the architecture is different because although the constituting elements aresimilar as for the individuals, the roles and tasks are different Again, the discussionabout the character of the architecture boils down to a discussion about internally orexternally defined Internal are chips and the cognitive architecture, whereas groups

of people and organizations can be dealt with externally

(Internal) Representations In cognitive science the conceptual framework to dealwith representations can be found in the approaches of classical symbol systems,connectionism, and situated action (Posner, 1989; Newell, 1990; Do¨lling, 1998;Smolensky, 1988; Jorna, 1990) The basic idea in classical symbol systems theory

is that humans as information processing systems have and use knowledge ing of representations and that thinking, reasoning, and problem solving consist ofmanipulations of these representations at a functional level of description Asystem that internally symbolizes the environment is said to have representations.Representations consist of sets of symbol structures on which operations are defined.Examples of representations are words, pictures, semantic nets, propositions, or tem-poral strings A representational system learns by means of chunking mechanismsand symbol transformations (Newell, 1990) A system is said to be autonomous

consist-or self-consist-organized if it can have a representation of its own position in the ment This means that the system has self-representation Connectionism and situ-ated action are attacks on missing elements within the classical symbol systemapproach Connectionism criticizes the neglect of the neurological substratewithin the symbols approach and defends the relevance of subsymbolic processing

environ-or parallel distributed processing Situated action criticizes the neglect of theenvironment within the symbol approach and emphasizes the role of actions,situatedness, and “being in the world.”

Mostly, plan execution takes place in the environment of an entity An entity thatmakes a plan for itself can of course misinterpret its position in the environment—for example, because it cannot represent its environment (e.g., the primitives of themodel of the future have not enough expressive power) or because it cannot mani-pulate its representation of the environment adequately Furthermore, an entity thatmakes a plan for others can additionally have this problem with respect to the

entities that must execute the plan Representations are also immediately relevant foranticipation A description of a future state in whatever symbol system or signsystem is the core of any discussion about anticipation.

Communication, Meaning, and Interpretation Communication means theexchange of information between different components Depending on whether

we are talking about internal or external information processing entities, cation means possibilities for and restrictions on the kinds of symbols or signs (thecodes) that are used for information exchange If we relate this to the aforemen-tioned discussion about representations, the various kinds of signs have differentconsequences Clearly, sign notations are more powerful (allow less ambiguity),but also more restricted than sign systems, which in turn are more powerful thanjust sign sets (Goodman, 1968; Jorna, 1990; Jorna and van Heusden, 1998) Unam-biguous communication requires sign notations (reducing ambiguity as much aspossible), but we know that all communication between humans is not in terms ofnotations If computers require sign notations and humans work with sign systems,then if the two have to communicate, they have to adjust to each other Untilrecently, most adjustments consist of humans using notations Now, interfaces aredesigned that allow computers to work with less powerful (allowing more ambi-guity) but more flexible sign systems This means that computers can now betterdeal with ambiguity For mental activities no explicitness (channels, codes, etc.)

communi-is necessary; for planning as an external activity, managing others in organizations,

an artificial intelligent actor that in its ultimate simplicity could be a chip In case

of a set of entities that are not by themselves a coherent unity—for example, viduals in an organization—various coordination mechanisms can be found, such

indi-as a hierarchy, a meta-plan, mutual adjustment, a market structure, and manyothers (Thompson, 1967; Gazendam, 1993) The important difference with singlehuman actors is that these coordination mechanisms are external and of coursewith direct access

Planning, Execution, and Control Making a plan, executing it, and monitoring theoutcomes in reality are valued differently in planning your own actions compared toplanning actions of others (i.e., organizational processes) Planning in organizationsusually is decoupled from the execution of the plan There are two main reasons whythe planner is someone other than the one who executes the plan First, planning is adifficult job that requires expertise and experience This is the organizational con-cept of task division Second, a planner must be able to weigh the interests ofmany parties Therefore, he must have knowledge about things that go beyond the

limits of the individual tasks that are planned The consequence of this decoupling isalmost always inflexibility with respect to adaptation For simple tasks such as doingerrands, the possible division in terms of subtasks may be interesting, but can in rea-lity be intertwined with flexible adaptation after unforeseen events If the controllingentity is itself a unity, discussions about transfer, communication, sign systems to dothe communication, and representations are almost trivial This does not make theplanning task itself simpler; it only prevents the explicitly formulated occurrence

of ambiguity, interpretation, and meaning variance

Our starting point in this book is that planning is always in essence about thesame thing: anticipating on the future and determining courses of action Theabove-discussed characteristics allow us to determine the similarities and dissimila-rities of the various planning approaches and perspectives This will become moreevident in the next section and later in the consecutive chapters

1.6 CONSOLIDATION: FROM DIFFERENT PLANNING

WORLDS TO DIFFERENT PLANNING WORDS

In this chapter, we proposed a generic model of planning Someone or somethingcreates a plan, and someone or something executes that plan Both are acting in

an environment, and the planning entity, planned entity, and environment have anumber of characteristics with which a planning situation can be described Thecharacteristics can be used for all planning fields and thereby they can be used tocompare planning approaches The following list summarizes the characteristics:

Planning Environment Different planning approaches have different environments.Even within approaches different environments can be distinguished Some charac-teristics are:

1 The predictability of the environment (Is the plan executed in a closed world

in which the presumptions will not change, or is it executed in an open world?)

2 Kinds of events that trigger planning:

a Time-based: For example, a plan needs to be made each week

b Event-based: A plan must be made after an event, for example a rush order

of making the plan—for example, when the plan must be available or howmuch people may be involved in making the plan Goals and constraints

can be about time, materials, remaining life span of the system, energy, thedegree of fault-tolerance, money, capacity usage, and so on Making theconstraints and goals explicit is often the hard part in planning.

Planned Entities The actions in the plan must be performed by someone or thing Several aspects are important:

some-1 Is the planned entity the same as the planning entity?

2 Does the plan deal with actions of individuals or actions that are performed bygroups of individuals?

3 Is the planned entity a natural entity (e.g., human) or an artificial entity (e.g., amachine, a robot, or a computer program)?

4 Does the planned entity possess intelligence itself ? That is, can it interpret theplan and change it if necessary?

5 What kind of constraints do the planned entities impose on the plan?

6 Does the planned entity use scarce resources that also have to be planned?

Planning Entities Someone or something must make the plan—that is, search foralternative plans and choose one Important aspects of the planning entity are asfollows:

1 Does the planning entity execute the plan itself or is the plan executed byothers?

2 Is it a natural entity (human) or is it an artificial entity (computer program)?

3 What kind of planning methods does it use?

4 What is the planning strategy? That is, how does it choose an appropriate ning method?

plan-5 What kind of information processing mechanism does it have?

6 What are the architectural components?

7 What kinds of representations does it use?

8 How does it communicate?

9 How does it coordinate with other planning entities and with planned entities?

Plan The plan itself is the specification of future actions

1 Horizon: What time span does the plan cover?

2 Frequency: How often is the plan created or adapted?

3 Level of detail: Does the plan need more detail in order to be executed? Doesthe executing entity have to fill in the details, or is the plan used as a templatefor another planner?

4 Structure: What is actually planned—for example, human actions, machines,time, locations, vehicles, movements, and so on

A plan that contains explicit temporal assignments on an interval or ratiolevel of measurement is usually called a schedule.

5 (Re)presentation: How is the plan represented or depicted? Does it specifythe end state, or does it provide a process description that leads to the endstate?

Planning Methods The planning method depicts the decision process of the ning entity A planning entity can have multiple planning methods to choosefrom Some generic issues with regard to planning methods are as follows:

plan-1 How does the planning entity deal with combinatorics, for example?

a Plan partitioning: Divide the plan in multiple subplans and treat thesubplans independently

b Multiresolutional planning: Make a plan with less detail (and less plexity), and use that plan as a template for a plan with more detail (at ahigher level of resolution)

com-c Learning: Use (and possibly adapt) a previously found solution for a blem that was equal or similar

pro-d Opportunistic planning: Apply the first feasible solution that is found out looking whether there are better solutions (e.g., when planning understrict time constraints)

with-2 How much does the use of a planning method cost? Methods can, for example,

be costly in terms of the information processing capacity that is needed, or inthe tools that are used, or in the throughput time that is needed

3 What is the starting point? For example:

a An empty plan

b An existing plan that must be supplemented

c An existing plan with errors that must be corrected

d A previous plan that can be used as a template

4 How are conflicts during the solution process solved (i.e., when the planningentity gets stuck)?

a Backtracking

b Repair

c Adjustment of constraints to make the plan valid

5 Does the method search for an optimal solution or state, or does it search for asatisfactory solution?

In the various parts of this book, we will refer to these generic characteristics inorder to be able to show common aspects of different planning approaches Thus,our presumption is that those different planning approaches are not as different asthey appear The structure of this book in which the planning approaches arediscussed will be described in the next section

1.7 STRUCTURE OF THIS BOOK

There are two parts in this book: Part A is theoretical and part B is practical Part Acontains 10 theoretical chapters The chapters are not meant to be introductions tothe respective research fields Rather, each chapter explores one or more issues inits research field in detail Thereby exposing different planning issues, languages,models, and methods The respective theoretical contributions will be introducedbriefly

Hommel (Chapter 2) gives a psychophysiological analysis of planning He showsthat planning and plan execution are very much interrelated by discussing threeplanning steps that humans take during an action Thus, his contribution dealswith humans that plan their own future actions However, the planning horizon isvery short, less than one second

Chapter 3, by Hoc, contains cognitive analyses of the planning behavior of humanoperators in dynamic environments where there is no clear-cut distinction betweenplan generation and plan execution and where the human operator does not fully con-trol the environment and hence faces uncertainty—for example, in industrial processcontrol and in anesthesiology In contrast to Hommel, Hoc uses a functional level ofdescription rather than a physiological one As with Hommel, Hoc is a typicalexample of an analysis of humans that plan their own future behavior

In Chapter 4, Jorna analyzes the behavior of 34 planners in three planningdomains (staff scheduling, production planning, and transportation planning) intwo countries (The Netherlands and Indonesia) Each planner had to solve asimple planning problem that was structurally the same for all planners but modifiedfor each domain so the terminology was always comprehensible The contribution ofJorna exemplifies traditional (cognitive) task analyses (e.g., see the contribution byHoc), but since the objects of study are planners and their problem-solving processes,this study deals with humans that plan for or take part in organizational processes.Gazendam (Chapter 5) takes a broader perspective on planning by looking atorganizations as multi-actor systems In his view, planning is a form of coordinationthat takes place by negotiation and coordination Thus, his contribution can beclassified as one in which humans plan organizational processes, be it thatGazendam does not study individuals but rather groups of individuals

Following Gazendam’s line of reasoning, one would expect theories and ologies that can be used to evaluate or even design the organization of the planningfunction in organizations In Chapter 6, however, McKay and Wiers note that mostplanning research looks at planning in isolation from other organizational functions,disregarding cognitive aspects of human planners and the organizational context inwhich human planners work As a starting point to further enhance our knowledge

method-on the role of planning in organizatimethod-ons, McKay and Wiers propose a frameworkwith which the interconnections between planning and other organizational func-tions can be depicted

Complementing the line of reasoning of McKay and Wiers, van Wezel looks atcomputer support of planning in organizations Noting that the essential problems incomputer support have not changed in the past 30 years, van Wezel proposes a

structure for reuse in scheduling system development, and a framework for usingalgorithms in scheduling support systems.

Kroon and Zuidwijk show in Chapter 8 how techniques from OperationsResearch are applied to a multitude of planning problems in the Netherlands Rail-ways (in Dutch: NS) Clearly, this is a typical example of artifacts that makeplans for organizational processes They also stipulate that plan generation isimportant but not enough for full planning support

Sauer (Chapter 9) applies techniques from Artificial Intelligence to productionscheduling He, therefore, deals with artificial planning actors that plan organiz-ational processes In particular, he looks at problems that occur when schedules

of different factories must be attuned

Similar to Sauer, Bowling et al (Chapter 10) also apply techniques from cial Intelligence to attune the actions of multiple actors Unlike Sauer, however, thekinds of actors they investigate are robots or simulated robots that on the one handplan for themselves but on the other hand have to cooperate with or work againstothers

Artifi-Meystel considers in Chapter 11 the planning process as part of the controlprocess He shows that a view on planning as a nested system of state-spacesearch processes can yield efficient computational algorithms that are similar tothose obtained by human planning strategies Thus, his work can be classified asartificial agents that plan for themselves

Table 1.2 provides an overview of the theoretical contributions

TABLE 1.2 Theoretical Book Chapters

Planning Entity¼ Executing

Entity

PlanningEntity = Executing Entity

analyses of planning(Hommel, Chapter 2)

† Planning in cognitivepsychology (Hoc,Chapter 3)

† Cognitive aspects ofplanning inorganizations(Jorna, Chapter 4)

† Planning inorganizational science(Gazendam, Chapter 5;McKay and Wiers,Chapter 6)Planning Entity 5 Artificial † Planning by robots

(Bowling et al.,Chapter 10)

† Planning in complexautonomous systems(Meystel, Chapter 11)

† Artificial Intelligence inproduction planning(Sauer, Chapter 9)

† Planning support andreuse (van Wezel,Chapter 7)

† Planning in operationsresearch (Kroon andZuidwijk, Chapter 8)

Part II of this book contains descriptions of different planning approaches thatare applied to one case study, namely the shunting of trains at a station of theNetherlands Railways Passenger trains that enter a station at the end of the daymust leave the following morning, often in another configuration This requiresthat the trains have to be separated in carriages, the carriages have to be parked atshunting tracks, and the following morning they have to be combined again to pas-senger trains This problem is not only challenging from a mathematical perspective,but also interesting from a (cognitive) task and organizational perspective, sincedozens of planners at the Dutch Railroad Company work on this problem Afterdescribing the shunting problem in Chapter 12, the respective case study chapterswill discuss how the problem was tackled from the different approaches (Table 1.3).

In the conclusions (Chapter 18), we assess the various planning approaches thatare discussed in this book with the framework that is depicted in this introductorychapter We will show that planning approaches are comparable and that oftendifferent planning words are used to describe the same phenomena In Chapter 18

we also propose a research agenda to make use of the generic elements of planningand scheduling

TABLE 1.3 Application-Oriented Chapters

PART I

THEORETICAL

INTRODUCTION TO CHAPTER 2

Hommel’s chapter describes a kind of planning that may not be planning at firstsight In Hommel’s studies, subjects get a task they have to carry out after somekind of event For example, subjects must press the left button when a green lightswitches on, and they must press the right button when a red light switches on.Hommel takes a cognitive neuropsychological point of view, where the planninghorizon is extremely short and where the planning frequency is so high that planningand execution seem very much integrated

Can we speak of planning here? When we observe the action behavior of ahuman, we cannot observe a planning process as such Planning and action arealmost simultaneous The planning process stays internal and is hard to analyzebecause the subject performs it subconsciously However, a robot may possess thesame characteristics The only difference might be that we know we have pro-grammed the robot’s behavior with some kind of planning algorithm, whereas theway in which a human plans his or her actions is largely unknown To settle thisfor Hommel, we will analyze how human actions score on the definition of planningdiscussed in the introduction There we stated that planning is a synthetic task inwhich an alternative from a large set of structurally similar alternatives has to bechosen based on constraints and goals

First, an action is the synthesis of a number of underlying activities, so ing the action is a synthetic task Second, the planning phase must be discernablefrom the action phase Although not apparent while observing human actions,Hommel shows on the basis of experiments that such a distinct planning phase

exists Third, planning means choosing from a large number of structurally similaralternatives On the one hand, choosing between the left or right button after seeing agreen or red light results in a very low number of structurally similar alternatives Onthe other hand, however, we should note that pressing the button is an experimentaltask for research purposes More involved actions such as riding a bicycle or driving

a car are much more complex, but essentially the same as choosing the button topress on the basis of a visual cue Fourth, human actions are subject to constraintsand goals when choosing one of the alternative courses of action For example, con-sider a ball that someone has to catch The movement that we want to make is lim-ited by constraints such as objects that block the movement, the flexibility andinertia of our muscles and bones, and the fact that the palm of our hand must betoward the ball at the end of the movement The goal is to maximize the chancethat we catch the ball, which can be a trade-off between the speed with which wemove our arm and the accuracy with which we move it

Overall, human actions do possess all the characteristics that make up planning

In his chapter, Hommel will extensively discuss issues such as the development ofplanning methods by infants, the way in which we focus our attention in actionplanning, and the hierarchy of control and movement that can be found in thestimulus – response cycle In the concluding chapter (18), we will compare humanactions planning with the other planning approaches, and there we will analyzewhy humans are able to make complicated planning decisions within a few milli-seconds, whereas making a planning decision with a computer can take hours and

a planning decision in an organization can take days