• CHAPTER 3 : ROBOT BEHAVIOR
• CHAPTER 4 : BEHAVIOR-BASED ARCHITECTURES

• Reactive Systems:
  • behaviors are building blocks
  • no abstract representation
  • animal models are used
  • modularity from Software engineering side
• Planning is often a waste of time
• Methods used for design of behaviors:
  • Ethologically guided
  • Situated-activity based : A good understanding of robot-environment interaction according to situations -> some activities are defined
    • More general case : Universal plans
    • Reactive Action Packages (RAPs) are not behavior based, they are task based
  • Experimentally driven design

Expression of Behaviors

• Stimulus response diagrams. Stimulus -> Behavior -> Response
• Functional notation
• Finite State Acceptor Diagrams
• Formal Methods:
  • Robot Schema : Combine advantage of functional notation and FSA methods into a single syntax
  • Situated Automata : Logical formalism, fundamental relationship on agent which has as a participant within its environment. goal reduction (Gapps).

Behavioral Encoding

• Functional map from stimulus to motor plane
• Degree of freedom (range of response)
• Behavioral encoding:
  • NULL
  • Discrete encoding -> ex: situated action, rule-based systems, Gapps. actions may be durative response
  • Continuous encoding -> infinite space of potential reactive -> potential fields methodology :
    • goals treated as attractors, obstacles are treated as repulsors
    • separate fields are combined
    • vulnerable to local minima - cyclic oscillation behavior
    • time to compute fields ( only field which is at robot location is computed ) can be very fast and parallelizable, only local computations.
      • Navigational Templates (NATs)
      • deformational zones for collision avoidance

Assembling Behaviors

• Emergency : Sum is greater than parts, interacting components
• control functions are well designed or defined
• BUT relationship the robot with its environment is difficult since
  • uncertain and dynamic environment
  • sensors are not precise
  • complexity of perception of the complex world
• Subsumption : Behavioral prioritization
• A formal notation is given which takes stimulus behavior functions, gain of behaviors as metrics and apply a central function.
• Behavioral coordination : competitive and cooperative methods
• Behavioral Assemblages are constructed for modularity and reusability and abstraction power.

CHAPTER 4 : BEHAVIOR-BASED ARCHITECTURES

What is a robotic architecture?

• Definition of robotic architecture term
• All the class of robotic architectures are equivalent in the sense of computability
• Evaluation criteria:
  • support for parallelism
  • hardware targetability
  • niche targetability
  • support for modularity
  • robustness
  • timeliness in development
  • runtime flexibility
  • performance effectiveness

Subsumption Architecture

• Rodney Brooks : purely reactive behavior based
• The attack of Rodney Brooks to traditional robotic approaches, reasons...
• Tasks are represented as layers in subsumption. Individual layers work on individual goals. Concurrency and asynchronously.
• Suppression and/or inhibition by other active behaviors.
• Behavior Language.
• Coordination : Incremental design (complex actions subsume simpler behaviors )
• Communication between layers is possible but restricted heavily.
• Design : Key issues are situatedness (sense)  and embodiment (physical).
• Evaluation :
  • positive : Hardware targetability, support for parallelism, niche targetability.
  • middle : Robustness, timeliness for development
  • negative : Run time flexibility, support for modularity

Motor Schema

• Arkin : Schema provide large grain modularity
• Arkin : Act concurrently
• Provide a set of behavioral primitives -> complex behaviors (assemblages) constructed.
• Continuous response encoding
• Cooperative coordination
• No hierarchy exists for coordination
• Strengths of behaviors determine robot's overall response

Schema-Based Behaviors

• Behaviors have internal parameters - flexibility
• Analogous to animal behaviors
• Action-oriented sensing (coupled perceptual-motor schema)
• Although entire field is illustrated, only a single vector needs to be computed are the robot's current location -> very fast.
• Coordination : Vector summation
• Noise schema can be inserted into system to deal with local minima problem - injection of randomness
• Avoid Past Schema can be applied into the system
• Evaluation :
  • positive : Support for parallelism, run-time flexibility, timeliness for development, support for modularity
  • middle : robustness
  • negative : Niche targetability, hardware re-targetability.
•