Transcript Robot Intelligence Technology Lab

EE788 Robot Cognition and Planning, Prof. J.-H. Kim
Ch. 1 Toward Intelligent Robots
From R. C. Arkin, Behavior-based Robotics, The MIT Press, 1998
Robot Intelligence Technology Lab..
1
Lecture Objectives

To understand what intelligent robots are.

To review the recent history that led to the development of
behavior-based robotic systems.

To learn and appreciate the wide spectrum of robot control
methods.
Robot Intelligence Technology Lab.
2
Contents
1. Toward intelligent robots
2. Precursors
2.1 Cybernetics
2.2 Artificial intelligence
2.3 Robotics
3. The spectrum of robot control
3.1 Deliberative/Hierarchical control
3.2 Reactive systems
4. Related issues
Robot Intelligence Technology Lab.
3
1. Toward intelligent robots

If we could create intelligent robots,



What should they be like?
 The robot’s physical structure (appearance)
 The robot’s performance (behavior)
What should they be able to do?
 Robots that need to move objects must be able to grasp them.
 Robots that must function at night need sensors capable of operation
under those conditions.
What is a robot?


“A robot is a re-programmable, multi-functional, manipulator designed
to move material, parts, tools, or specialized devices through variable
programmed motions for the performance of a variety of tasks”
(Robotics Industry Association, Jablonski and Posey 1985)
(Considering mobile robots) “The intelligent connection of perception
to action” (Brady 1985)
Robot Intelligence Technology Lab.
4
1. Toward intelligent robots

Our working definition


An intelligent robot is a machine able to extract information
from its environment and use knowledge about its world
to move safely in a meaningful and purposive manner.
Various robots



Anthropomorphic robots
Animal-like robots
Unmanned vehicles

Different robots







Robot Intelligence Technology Lab.
The size
The materials
The way they are joined together
The actuators
The types of sensing systems
The locomotion system
The onboard computer systems
5
1. Toward intelligent robots (cont’d)

This lecture focuses on the performance and behavioral aspects
of robotics and the design of control systems that allow them to
perform the way we would like:




the basis and organization of behavior
the related roles of knowledge and perception,
learning and adaptation, and
teamwork.

How do we realize the goal of intelligent robotic behavior?

What basic science and technology is needed to achieve this goal?
Robot Intelligence Technology Lab.
6
2. Precursors

2.1 Cybernetics
The significant history associated with the origins of modern
behavior-based robotics:



Cybernetics
Artificial intelligence
Robotics
 The development of cybernetics

In the late 1940s, Norbert Wiener

Control theory + information science + biology

To explain the common principles of control and communication
in both animals and machines
Robot Intelligence Technology Lab.
7
2.1 Cybernetics

Ashby and Wiener (1952)

The view of an organism as a machine by using the mathematics
developed for feedback control systems to express natural
behavior
In 1953, Machina Speculatrix
(Grey Walter’s tortoise)
Robot Intelligence Technology Lab.
8
2.1 Cybernetics (cont’d)

The tortoise

Constructed as an analog device

Two sensors (a photocell and a contact sensor)

Two actuators

Two “nerve cells” or vacuum tubes

Behaviors
 Seeking light
 Head toward weak light
 Back away from bright light
 Turn and push
 Recharge battery
Robot Intelligence Technology Lab.
9
2.1 Cybernetics (cont’d)

Some of the principles in the design of the tortoise:

Parsimony: Simple is better.

Exploration or speculation:
 The system never remains still except when feeding (recharging).

Attraction (positive tropism):
 The system is motivated to move towards some environmental object.

Aversion (negative tropism):
 The system moves away from certain negative stimuli, for example,
avoiding heavy obstacles and slopes.

Discernment:
 The system has the ability to distinguish between productive and
unproductive behavior, adaptation itself to the situation at hand.
Robot Intelligence Technology Lab.
10
2.1 Cybernetics (cont’d)

Braitenberg creatures

Using inhibitory and excitatory influences, directly coupling the sensors
to the motors (1984)

True robots (experiments)
 1991, MIT’s Media Lab
 Using LEGO bricks
 Twelve autonomous creature-vehicles using Braitenberg’s principles
 Including a timid shadow seeker, an indecisive shadow-edge finder,
a paranoid shadow-fearing robot, a dogged obstacle avoider, an insecure
wall follower, and a driven light seeker
Robot Intelligence Technology Lab.
11
2.1 Cybernetics (cont’d)

Braitenberg Vehicles
Robot Intelligence Technology Lab.
12
2.2 Artificial Intelligence

The birth of AI

The Dartmouth Summer Research Conference (1955)

In the original proposal (McCarthy)
 An intelligent machine “would tend to build up within itself
an abstract model of the environment in which it is placed.
If it were given a problem it could first explore solutions
within the internal abstract model of the environment and
then attempt external experiments.”
Robot Intelligence Technology Lab.
13
2.2 Artificial Intelligence (cont’d)


Two important characteristics of the classical AI

The ability to represent hierarchical structure by abstraction

The use of “strong” knowledge that employs explicit symbolic
representational assertions about the world
AI’s influence on robotics

In the idea that knowledge and knowledge representation are
central to intelligence, and that robotics was no exception.
Robot Intelligence Technology Lab.
14
2.2 Artificial Intelligence (cont’d)

Behavior-based robotics system reacted against AI traditions.




“Planning is just a way of avoiding figuring out what to do next.” Brooks 87
The notion of sensing and acting within the environment started to take
preeminence in AI-related robotics research over the previous focus on
knowledge representation and planning.
Distributed AI (DAI) paralleled these developments.
 Multiple competing or cooperating processes (initially demons and later
agents) are capable of generating coherent behavior.
Multi-agent system: Minsky’s Society of Mind Theory (Minsky’86)

The basis for all intelligence, although each agent is as simple as it can be,
through the coordinated and concerted interaction between these simple
agents, highly complex intelligence can emerge.
Robot Intelligence Technology Lab.
15
2.3 Robotics

Mainstream robotics have by necessity
generally been more concerned with
perception and action.

To conduct robotics research, robots are
needed. Those who only work with
simulations often ignore this seemingly
obvious point.
Shakey,
One of the first mobile roots, Nilsson 1969,
Stanford Research Institute
Two stepper motors, a vidicon television camera,
optical range finder, bump sensors
Robot Intelligence Technology Lab.
16
2.3 Robotics (cont’d)
Stanford Cart, 1977
Stereo vision as a mean for navigation,
quit slow (visual processing)
HILARE, 1984, LAAS, France
3 wheels (two drive and one caster), 400kg, video
camera, 14 ultrasonic sensors, laser range finder
Still being used for experimentation
Robot Intelligence Technology Lab.
17
2.3 Robotics (cont’d)
CMU Rover, 1983, Moravec, CMU
A smaller, cylindrical robot with 3 independently
powered and steered wheel pairs,
Camera mounted on a pan/tilt mechanism,
Infrared and ultrasonic sensors

Robot Intelligence Technology Lab.
These and other robotic precursors set the
stage for the advances and controversies
to come as behavior-based robotic
systems appeared in the mid-1980s.
18
3. The spectrum of robot control

Robot control system spectrum
Robot Intelligence Technology Lab.
19
Deliberative reasoning systems

Hierarchical in structure with a clearly identifiable subdivision
of functionality, similar to the organization of commercial
business or military command.

Communication and control occurs in a predictable and
predetermined manner, flowing up and down the hierarchy,
with little lateral movement if any.

Higher levels in the hierarchy provide subgoals for lower
subordinate levels.

Planning scope: At the lower levels, time requirements are
shorter and spatial considerations are more local.

They rely heavily on symbolic representation world models.
Robot Intelligence Technology Lab.
20
3.1 Deliberative/Hierarchical control

Albus’s hierarchical intelligent control system

All layers are joined by a global memory through which representational
knowledge is shared.
Robot Intelligence Technology Lab.
21
3.1 Deliberative/Hierarchical control (cont’d)

NASREM architecture

A standard architecture
(US Government, ’87)

Standard reference model
for Telerobot Control
System Architecture
Robot Intelligence Technology Lab.
22
3.1 Deliberative/Hierarchical control (cont’d)

Mobile Robot Control System:
Nested hierarchical intelligent controller
Robot Intelligence Technology Lab.
23
3.1 Deliberative/Hierarchical control (cont’d)

Logical model of hierarchical intelligent robot: Lefebvre & Saridis, ’92

IPDI Principle:
Increasing Precision,
Decreasing Intelligence

Organization level:
High-level planning and
reasoning

Coordination level:
Integration across various
hardware subsystems

Execution level:
Basic control and hardware
Robot Intelligence Technology Lab.
24
3.2 Reactive systems

Reactive control is a technique for tightly coupling perception
and action, typically in the context of motor behaviors, to produce
timely robotic response in dynamic and unstructured worlds.

An individual behavior:
A stimulus/response pair for a given environmental setting that is
modulated by attention and determined by intention.

Attention:
Prioritizes tasks and focuses sensory resources and is determined
by the current environmental context.

Intention:
Determines which set of behaviors should be active based on the
robotic agent’s internal goals and objectives.
Robot Intelligence Technology Lab.
25
3.2 Reactive systems

Overt or emergent behavior:


The global behavior of the robot or organism as a consequence of
the interaction of the active individual behaviors.
Reflexive (reactive) behavior:

Generated by hardwired reactive behaviors with sensor-effector arcs.
Key Aspects of this Behavior-based Methodology, Brooks, 1991
 Situatedness:


Embodiment:


The robot is an entity situated and surrounded by the real world.
A robot has a physical presence.
Emergence:

Intelligence arises from the interactions of the robotic agent with its
environment.
Robot Intelligence Technology Lab.
26
4. Related issues



Grounding in reality
 AI: a symbol grounding problem
“Building robots that are situated in the world crystallizes
the hard issues.” (Flynn and Brooks ’89)
Ecological dynamics
 In nature, evolutionary processes shape agents to fit their
ecological niche; these time scales unfortunately are not
available to the practicing roboticists. Adaptation, however
can be crucially important.
Scalability
 Although behavior-based methods are clearly well suited
for low-level tasks requiring the competence of creatures
such as insects, it has been unclear whether they would
scale to conform to human-level intelligence.
Robot Intelligence Technology Lab.
27