Agent Oriented theory of Human Activity

1
Agent Oriented theory of Human Activity

• Thesis Craig Rindt (Chapter 3)

2
The general Aim

• Apply Agent-based modeling techniques to general
  activity systems theory to model human travel
  behavior.
• What is Activity Systems theory?
• Peoples travel behavior can be understood in the
  context of activities they want to do.

3
Definitions

• Activity
• Episode Discrete event occurring over time.
• Trajectory actual behavior over time.
• Pattern Analytical description of trajectory in
  time and space.
• Action space Set of actions that are feasibly
  reached over space and time.
• Calendars demands to engage in activities
• Programs Agenda of activities that must be
  performed
• Schedules Planned trajectory that an individual
  decides.

4
Various theories on Activity systems analysis

• Theory 1(Constraints)
• States that Human behavior is a constrained
  trajectory through time and space.
• Types of Constraints
• Capability constraints arising due to physical
  limitations
• Coupling constraints arising from interactions
• Authority constraints define personal control of
  resources e.g. I cannot shop at a store if it is
  closed

5
Theory2 (Motivation)

• Concentrates on propensity factors that drives
  humans to do stuff.
• Not articulated properly and a lot of different
  cases exist.
• Main Idea Human behavior in space is
  characterized by the motivation to participate in
  various activities.

6
Theory 3 and 4

• Balancing Motivation and Constraints
• Neither all activities nor all constraints are
  equal in the eyes of the actor or a weighted
  theory.
• Adaptation
• Individual is situated in an environment that
  both motivates and constrains his behavior.

7
(No Transcript)
8
(No Transcript)
9
Idea in the thesis

• Combine the theories just described with Agent
  based modeling philosophy.
• Agent-based View
• A Human-agent occupies a universe filled with
  other agents.
• Agents knowledge gained solely through sensors.
• Effectors
• Achieve GOALS by interaction with other agents.

10
(No Transcript)
11
(No Transcript)
12
Activity as Interaction

• The agent-based view states that the behavior of
  an agent depends upon the interaction it has with
  the other agents.
• gt Activity Interaction
• Thus, Human Activity can be viewed as both
  mechanism of constraint and source of motivation.

13
Defining the Human Agent

• Some Assumptions Assume you can synthesize a
  population of agents in an urban environment by
  using some techniques.
• Such a technique also specifies the social
  structure and things like physical proximity.
• Now, we seek to produce for each agent, the
  following time-varying vector
• Y(t) XL(t),XC(t),XA(t)
• XL, XC and XA stand for location, social impact
  and interaction respectively

14
Representing dynamics

• Y(t)XL(t),XC(t),XA(t)
• f(XL(t-1),XA(t-1)),f(XC(t-1),XA(t-1)),f(R(t-1
  ),P(t-1))
• R(t) Resources available to the agent at time t.
• P(t) Agents plan.

15
Specifying Resources or Interfaces

• View
• The resources available effectively define the
  channels upon which an individual can interact
  with the environment to engage in an activity.
• Each agent therefore has an interface that it
  presents to other agents which represents the
  types of interactions it can have.
• R(t) f(XL(t),XC(t),L(t), T(t),C(t))
• L(t), T(t) and C(t) are the land-use system, the
  transportation system and the socio-cultural
  system respectively.

16
So What?

• The goal of activity and travel forecasting is to
  predict this trajectory Y over time. (Economic
  models)
• The goal of transportation science is to describe
  and understand how human behavior produces the
  trajectory. (Learning problem)
• The behavior is dependent on the plan P
• P(t)f (P(t-1), XL,XC,E(t))
• where E(t) (L(t),T(t),C(t)) is the environment.

17
Specifying Agent internals

• Assume that the environment is enumerable E
  (e1,e2,).
• The Agent has only partial knowledge of the world
  and so it considers the environment as R
  (r1,r2,r3.).
• ri is a subset of E.
• Define two functions,
• f E ? M (Sensory input to form messages)
• f M ? R (messages encoded to develop a
  perspective of the world)

18
Action-space and Agents view of the Action space

• Same as Sensory input.
• Available Actions S (s1,s2,.)
• Agents view A(a1,a2.)
• ai is a subset of S.
• To summarize E and R define the possible states
  of the objective world and the agent's ability to
  perceive that world.
• S and A define the universe of possible actions
  and the agent's subjective knowledge of them.

19
Completing the Agent description

• Interpretation.
• attribute a causal sense to the perceived world
  according to the agent's experience
• f H ? I (Historical information to
  Interpretation)
• Decisions
• f I ? A (Interpretation to activities)
• Assessing response for Actions through sensors.
• F E x S ? E

20
Completing the Agent description

• Agents utility functions
• U Z(I,B) where U is a Real number.
• Z can be interpreted as the agent's utility
  function, with B defining the utility weights and
  I defining the perceived values of the relevant
  attributes.
• Pay-off functions.
• f I X A ? U
• which is a mapping from the universe of possible
  interpretation-action combinations to some payoff
  measure in a range of utilities U.

21
Learning

• 4 Levels
• Learning about the states of the world (improving
  perception)
• Increase or decrease states in R.
• Learning About the Opportunity Space
• Increase or decrease states in A.
• Learning About Interpretations of Historical
  Trajectories
• Learning About the Decision Rules

22
Summary

• The focal agent is the human being, who is
  relationally situated to physical and social
  hierarchies that both motivate and constrain his
  behavior.
• This behavior is limited to interactions with
  other agents (people, institutions etc) from
  which the person derives some environmental
  payoff.
• Interactions can be conceived as a negotiation
  process which is the next chapter in the thesis.

23
Chapter 4

• The Micro-simulation Kernel

24
Introduction

• Recap Human Activity involves the interactive
  exchange of resources between individuals.
• View this as Negotiation
• Negotiation is driven by physical and social
  laws.
• Develop model according to this criteria and also
  try to reduce its complexity.

25
(No Transcript)
26
Design of Activity Negotiated Kernel

• Use Distributed Problem solving architecture
  (DPS)
• Model a urban system as a multi-agent system
  where agents represent people, institutions and
  places.
• Use an event-driven discrete model because the
  number of activities is not likely to exceed 50.

27
DPS and Contract Net Protocol (CNP)

• How to view DPS as negotiation based protocol
  (Davis and Smith 1983)---AnsCNP.
• Problems in DPS
• Each agent has an incomplete local knowledge
• Synchronize behavior so that agents dont
  interfere with actions of other agents.

28
(No Transcript)
29
Activity engagement as DPS

• Turn the CNP argument on its head.
• Activity engagement is the process used to solve
  the problem of activity completion.
• Problems
• No centralized problem solver in human activity
  negotiation.
• Solution
• View the task manager as an abstraction that
  represents the logic representing how physical
  and social constraints affect the laws of the
  environment.

30
(No Transcript)
31
Additions to CNP for Travel Domain

• Contracts involving multiple agent
• Non-binding contracts
• Terminate some activity at will.
• Binding Contracts
• E.g. Travel activities using Rail
• Simultaneous Activities

32
(No Transcript)
33
(No Transcript)
34
(No Transcript)
35
Summary