AI Robotics

Introduction to AI Robotics


I Robotic Paradigms

  1. From Teleoperation to Autonomy
  2. The Hierarchical Paradigm
  3. Biological Foundations of the Research Paradigm
  4. The Reactive Paradigm
  5. Designing a Reactive Implementation
  6. Common Sensing Techniques for Reactive Robots
  7. The Hybrid Deliberative/Reactive Paradigm
  8. Multi-agents

II Navigation

  1. Topological Path Planning
  2. Metric Path Planning
  3. Localization and Map Making
  4. On the Horizon

1. Robotic Paradigms

Overview of the Three Paradigms

  • The Hierarchical Paradigm
    • 1967-1990
    • operates in a top-down fashion, heavy on planing
    • SENSE, PLAN, ACT (S, P, A)
    • the input to a ACT will always be the result of a PLAN
    • due to the frame problem and the need for a closed world assumption
  • The Reactive Paradigm
    • 1988-1992~
    • to investigate biology and cognitive psychology in order to examine living exemplars of intelligence
    • SENSE-ACT (S-A)
    • the input to an ACT will always be the direct output o a sensor, SENSE
    • SENSE-ACT couplings that are concurrent processes, called behaviors
    • serves as the basis for the Hybrid Deliberative/Reactive Paradigm
  • The Hybrid Deliberative/Reactive Paradigm
    • 1992~
    • the robot first plans (deliberates) how to best decompose a task into subtasks and then what are the suitable behaviors to accomplish each subtask, etc. Then the behaviors start executing as per the Reactive Paradigm.
    • PLAN, SENSE-ACT (P, S-A)

3 Biological Foundations of the Reactive Paradigm

3.2 What Are Animal Behaviors?

  • A behavior is a mapping of sensory inputs to a pattern of motor actions which then are used to achieve a task.
  • Scientists who study animal behaviors are called ethologists.
  • Behaviors can be divided into three broad categories
    • Reflexive behaviors are stimulus-response (S-R)
      • hardwired
    • Reactive behaviors are learned, and then consolidated to where they can be executed without conscious thought.
      • muscle memory
    • Conscious behaviors are deliberative
      • assembling a robot kit, stringing together previously developed behaviors, etc.
  • In ethology, reactive behavior means learned behaviors or a skill; in robotics, it connotes a reflexive behavior.

3.2.1 Reflexive behaviors

Reflexive behaviors can be further divided into three categories:

  1. reflexes: where the response lasts only as long as the stimulus, and is proportional to the intensity of the stimulus.
  2. taxes: where the response is to move to a particular orientation.
  3. fixed-action patterns: where the response continues for a longer duration than the stimulus.

3.3 Coordination and Control of Behaviors

The four ways to acquire a behavior are:

  1. Innate
  2. Sequence of Innate behaviors
  3. Innate with Memory
  4. Learn

An important lesson that can be extracted from Lorenz and Tinbergen's work is that the internal state and/or motivation of an agent may play a role in releasing a behavior.

3.3.1 Innate releasing mechanisms

  • Lorenz and Tinbergen attempted to clarify their work in how behaviors are coordinated and controlled by giving it a special name innate releasing mechanisms (IRM).
  • An IRM presupposes that there is a specific stimulus (either internal or external) which releases, or triggers, the stereotypical pattern of action.
  • A releaser is a latch or a Boolean variable that has to be set.

3.3.2 Concurrent behaviors

  • Equilibrium
    • the behaviors seem to balance each other out
  • Dominance of one
    • winner take all
  • Cancellation
    • the behaviors cancel each other out

3.4 Perception in Behaviors

3.4.1 Action-perception cycle

When an agent acts, it interacts with its environment because it is situated in that environment; it is an integral part of the environment. So as it acts, it changes things or how it perceives it (e.g., move to a new viewpoint, trigger a rock slide, etc.). Therefore the agent's perception of the world is modified.

3.4.4 Neisser: Two perceptual systems

  1. direct perception
  2. recognition

3.5 Schema Theory

3.6 Principles and Issues in Transferring Insights to Robots

Several unresolved issue:

  • How to resolve conflicts between concurrent behaviors?
  • When are explicit knowledge representations and memory necessary?
  • How to set up and/or learn new sequences of behaviors?

4 The Reactive Paradigm

4.2 Attributes of Reactive Paradigm

4.2.1 Characteristics and connotations of reactive behaviors

The five characteristics of almost all architectures that follow the Reactive Paradigm are:

  1. Robots are situated agents operating in an ecological niche.
  2. Behaviors serve as the basic building blocks for robotic actions, and the overall behavior of the robot is emergent.
  3. Only local, behavior-specific sensing is permitted.
  4. These systems inherently follow good software design principles.
  5. Animal models of behavior are often cited as a basis for these systems or a particular behavior.

4.3 Subsumption Architecture

  • A behavior is a network of sensing and acting modules which accomplish a task.
  • The module area augmented finite state machines AFSM, or finite state machines which have registers, timers, and other enhancements to permit them to be interfaced with other modules.

Four interesting aspects of subsumption in terms of releasing and control are:

  1. Layers of competence
  2. Layers can subsume lower layers
  3. No internal state
  4. Taskable

4.3.2 Subsumption summary

  • Subsumption has a loose definition of behavior as a tight coupling of sensing and acting.
  • Higher layers may subsume and inhibit behaviors in lower layers, but behaviors in lower layers are never rewritten or replaced.
  • The design of layers and component behaviors for a subsumption implementation, as with behavioral design, is hard; it is more of an art than a science
  • There is nothing resembling a STRIPS-like plan in subsumption.
  • Subsumption solves the frame problem by eliminating the need to model the world.
  • Perception ins largely direct, using affordances. The releaser for a behavior is almost always the percept for guiding the motor schema.
  • Perception is ego-centric and distributed.

4.4 Potential Fields Methodologies

6 Common Sensing Techniques for Reactive Robots


  • Sensor / Transducer
  • Passive Sensor / Active Sensor
  • Active Sensing

6.1.1 Logical sensors
A logical sensor is a unit of sensing or module that supplies a particular percept.

6.2 Behavioral Sensor Fusion
Sensor fusion is a broad term used for any process that combines information from multiple sensors into a single percept.

The motivation for sensor fusion stems from three basic combinations of sensors:

  • redundant (or competing)
  • complementary
  • coordinated

Sensor fusion can be incorporated into behaviors through

  • Sensor fission
  • Action-oriented sensor fusion
  • sensor fashion

7 The Hybrid Deliberative/Reactive Paradigm

7.2 Attributes of Hybrid Paradigm

  • In the behaviors, sensing remains as it was for the Reactive Paradigm: local and behavior specific. But planning and deliberation requires global word models.
  • The organization of the SENSE, PLAN, ACT primitives in the Hybrid Paradigm is conceptually divided into a reactive (or reactor) portion and a deliberation (or deliberator) portion.

7.3 Architectural Aspects

  • How does the architecture distinguish between reaction and deliberation?
  • How does it organize responsibilities in the deliberative portion?
  • How does the overall behavior emerge?

7.3.1 Common components of hybrid architectures

  • Sequencer
    • generates the set of behaviors to use in order to accomplish a subtask, and determines any sequences and activation conditions
  • Resource Manager
    • allocates resources to behaviors, including selecting from libraries of schemas.
  • Cartographer
    • is responsible for creating, storing, and maintaining map or spatial information, plus methods for accessing the data
  • Mission Planner
    • interacts with human, operationalizes the commands into robot terms, and constructs a mission plan
  • Performance Monitoring and Problem Solving agent
    • allows the robot to notice if it is making progress or not

7.3.2 Styles of hybrid architectures

  • Managerial styles
    • focus on subdividing the deliberative portion into layers based on the scope of control, or managerial responsibility, of each deliberative function.
  • State hierarchies
    • use the knowledge of the robot's state to distinguish between reactive and deliberative activities
  • Model-oriented styles
    • characterized by behaviors that have access to portions of a world model.
Unless otherwise stated, the content of this page is licensed under Creative Commons Attribution-ShareAlike 3.0 License