Top-Down or Bottom-Up?

If you have come this far, that means you have completed three virtual robotics labs: one top-down and two bottom-up. You are now in a good position to reflect on the significance of these basic methods for producing "intelligent" behavior. You are to think carefully about the following questions and give your best answers.

NOTE: A Word document (robot.doc) with the questions is available HERE -- so that you can easily print the questions and/or write your answers on a computer.


1. What kinds of tasks are classic "top-down" approaches good for? What kinds of tasks would be problematic for this approach?

2. What kinds of tasks are behavior-based approaches good for? What kinds of tasks would be problematic for this approach?

3. Explain why you think that the right collection of very simple behaviors -- each one rather "stupid" by itself -- can together be combined to produce very "intelligent" behavior?

4. Why do top-down robots "hit a wall" (that is, reach a limit to their abilities) when the environment they are operating in gets too complex?

5. If you were going to design a subumption architecture to enable a robot to do     (fill in the blank)      what kinds of behaviors would you need and how would you design the architecture?

In the "blank" fill in each of the following:

a. Drive a car

b. Cook Pasta

C. Walk down a busy sidewalk

6. Why is a top-down robot, controlled by a central AI program, better at speaking a language than a bottom-up robot?

7. Human beings are the most intelligent things that we have found anywhere in the universe. Do you think human intelligence is primarily from a top-down architecture? A bottom-up architecture? Or some combination of the two? And most important -- whatever answer you gave, what features of human abilities and human intelligence lead you to give the answer that you did? Explain in detail.