1. Based on slides by Gal A. Kaminka
Introduction to Robots and Multi-Robot Systems Agents in Physical and Virtual Environments
Intro to AI, lesson 10
Based on slides by Gal A. Kaminka
and by Robin Murphy

2. Some examples of robots
Some examples of robots
courtesy of Honda
courtesy of MIT AI Lab

3. Less Famous Cousins at WTC
Inuktun microTracks
½ iRobot PackBot

4. What is a robot?
Give me a few examples.
Is a rock a robot?
Robot

5. A toy spring car can move and act.
What is a robot?
A toy spring car can move and act.
a robot can sense.
Robot
Actuators
(Effectors)

6. A sorting algorithm senses and acts.
What is a robot?
A sorting algorithm senses and acts.
a robot is persistent.
Robot
Actuators
(Effectors)
Sensors

7. What is a robot? a robot is situated in an environment.
What about a remote alarm?
a robot is situated in an environment.
Robot
Actuators
(Effectors)
Sensors

8. We’re missing something here.
What is a robot?
We’re missing something here.
a robot is responsive.
Robot
Actuators
(Effectors)
Sensors
Environment

9. We’re missing something here.
What is a robot?
We’re missing something here.
a robot is responsive.
Robot
Actuators
(Effectors)
Process
Sensors
Environment

10. Here’s what we have so far
Robots:
Are persistent with respect to their environment
Sense and act
Sense/act within the same environment (situated)
Respond to senses using action
Robot
Environment

11. Here’s what we have so far
Robots:
Are persistent with respect to their environment
Sense and act
Sense/act within the same environment (situated)
Respond to senses using action
These characteristic are true for agents, not just robots
Robot
Environment

12. Another definition Mechanical creature which can function autonomously
Capek 1921: R.U.R
Mechanical= built, constructed
Creature= think of it as an entity with its own motivation, decision making processes
Function autonomously= can sense, act, maybe even reason; doesn’t just do the same thing over and over like automation
Physically situated, but now software agents or bots

13. Why investigate robots? (1)
Because we want to understand how to build them.
So that they do things for us.
So that we can do other things instead.
In other words,
We are studying robotics because we are lazy.

14. Why investigate robots? (2)
Dirty, Dangerous, and Dull Tasks
Better Than Bio
Robots at WTC…
voids smaller than person could enter
voids on fire or oxygen depleted
Principles from robotics influenced AI community
Combines programming, networks, operating systems, algorithms, … everything about CS into a system (the ultimate software engineering project)
Void:1’x2.5’x60’
Void on fire

15. Minimally Invasive Surgery Spinal Fusion with Mazor’s SpineAssist

16. Mazor’s SpineAssist Surgical Robot

17. The Agent/Environment/Task Framework
Robot
Environment
Task
We want the robot to do tasks for us (or for itself)
Therefore, it must take a task into account

18. Problems with physical environments
Agents are embodied
Part of the environment is their own body
Sensing and acting with uncertainty
Slippery grips, sensing is inaccurate
Environment is dynamic, changes even without robot ….
We will talk more about environments later, but first….

19. A Taxonomy of Environments
There are a number of characteristic dimensions:
Dynamic vs. static
Accessible vs. inaccessible
transparent vs. translucent
Deterministic vs. non-deterministic
Discrete vs. continuous
…..

20. Is the agent only cause of change in the environment?
Dynamic vs. Static
Dynamic:
Environment changes even if agent takes no action
Static:
Environment does not change until agent takes action
Key question:
Is the agent only cause of change in the environment?
Physical environment is dynamic
Wind, other agents, continuous mechanical forces

21. Accessible vs. Inaccessible
Accessible (transparent):
Agent can sense everything and anything. Nothing is hidden.
Inaccessible (translucent):
Agent can only sense part of the environment.
Some features of the environment are hidden.
Key question:
What can the agent sense about the environment?
Physical environments typically inaccessible:
Cannot see behind you, nor over long distances, nor inside people.

22. If agent takes action, is it sure of the outcome?
Determinism
Deterministic:
An action results in a completely predictable change
Non-deterministic:
An action can result in one of a range of possible changes
Uncertainty in the result
Key question:
If agent takes action, is it sure of the outcome?
Physical environment is non-deterministic:
Slippery grasp, coin-flips, gambling

23. Discrete or continuous?
Actions or senses are clearly separated, limited number
Continuous:
Infinite possible values within a range
Note:
Different from discrete/continuous senses and actions
Physical environments are continuous

24. A Taxonomy of Environments
There are a number of characteristic dimensions:
Dynamic vs. static
Accessible vs. non-accessible
transparent vs. translucent
Deterministic vs. non-deterministic
Discrete vs. continuous
Open question: Quantifying the above

25. The Agent/Environment/Task Framework
Robot
Environment
Task
Given environment and task,
how do we build a robot that carries out the task?

26. Agents and Environments
Many different environments can exist
Different techniques are used with different environments
We focus on techniques used in physical environments

27. Agent Control
In principle, our view is of an agent with three components:
Effectors/actuators
Sensors
Think
This view is sometimes referred to as sense-think-act cycle
But this can be misleading: not necessarily so sequential
Think
Sense
Act
Robot
Environment

28. Three components, three challenges*
The action selection problem:
Given task/goals, how to select the next action(s)
The sensor planning problem:
Given task/goals, how to use sensors
The pose planning problem:
Given needed target body position, how to get there
Think
Sense
Act
Robot
Environment

29. Three components, three challenges*
The sensor planning problem:
Which sensors to use? When?
How to integrate their information (sensor fusion)?
How to overcome uncertainty in their readings?
May depend on what think is thinking, and may need to influence what action to take
Think
Sense
Act
Robot
Environment

30. Three components, three challenges*
The pose planning problem:
Which (combination of) actuators to use to achieve pose?
What trajectory should they take?
How to compensate for actuation uncertainty?
May depend on what think is thinking, and may need to depend what sense reads, and needs
Think
Sense
Act
Robot
Environment

31. Three components, three challenges*
The action-selection problem
How to select action in real-time?
How to select action that is good for task/goal?
How to integrate competing needs of different subtasks?
Depends on the capabilities of sense and act
Think
Sense
Act
Robot
Environment

32. Robotics is a highly inter-disciplinary field.
Three challenges
These three challenges are highly coupled
Not easy to separate them out.
Many systems/techniques provide integrated solutions
Multiple levels at which can be addressed:
hardware, control, software, …
Example: better vision by blurring camera
Example: using probabilistic inference to handle uncertainty
Example: sensor placement affects foraging behavior
Robotics is a highly inter-disciplinary field.

33. Empirical research As you can see, these are complex concepts
Many of problems/solutions affect each other in very subtle ways
Physical environments very uncertain, unpredictable
Difficult to predict system behavior from analysis
Cannot just browse at the algorithms and hardware involved
Use empirical research methods in investigations

34. Empirical research Experiment design issues:
Study system with and without proposed techniques
Compare performance of many systems
Compare performance across different environments or tasks
Faces generality problems in drawing conclusions
Tied to the actual challenges of the real world:

35. Simulations Significance issues: Simulation is very useful here
Run many experiments, draw statistical conclusions
Simulation is very useful here
Many roboticists frown at simulations
Simulation and virtual environment are not same thing

36. Science and Scientists
Scruffies and Neaties
The revolution of 86: Plans are not enough!

37. The Sense-Think-Act Cycle: What's in Think (for scruffies) in late 80's?
No need to Think: If sensors read X, then do Y
Reactive Camp (Brooks 1986, Schoppers 1987)
Limited thinking: Behavior-based control
Behaviors may have state, memory, procedures
Arkin, Firby (1986), Maes, ...
Deep thinking: integrated planning, monitoring
e.g., IPEM (1988)
Hybrid architectures (e.g., Gat 1992)

38. The Sense-Think-Act Cycle: What's in Think (for neaties) in late 80's?
"The Old View"
Plans as sequences of actions for execution
Plans as mental attitudes (Pollack 1992)
Plans as recipes: Some get executed, some just known
BDI: Belief-Desire-Intention (approximately):
Belief: What the agent knows
Desire: What the agents ideally wants to see happening
Intention: What the agents actually acts towards
Commitments

39. Subjective
An Historical Perspective on Teamwork: From a Single Agent to Multiple Agents
Scruffiness
Neatness
Time
Reactive-Plans,
Architectures
'86
Mental Attitudes,
Belief, Desire, Intention (BDI)
Plans as Attitude
Integrating
Planning, Execution,
Monitoring,
Re-Planning,
Architectures
Behavior-Based Architectures
'90
'96

40. Agent Teams Are Everywhere: Teamwork is Important
Nature
Formations, flocking, pack hunting, software development
Robotic nature imitations, explorations, soccer
Internet, Intranets
Routing, distributed applications, groupware
Workflow, cooperating information agents
Virtual environments for training, simulations
Human-computer interactions

41. Main research problems
Task-Specific Teamwork:
Foraging
Coverage
Mapping
Coordinated movement
Patrolling
Human supervision ….
General Teamwork:
Architecture
Flexible teamwork
Allocation
Learning, adaptation
Monitoring
Fault diagnosis
Maintenance ….