Page 77 - Modern Robotics Building Versatile Macines

REAL-WORLD ROBOTS 57

about costs, you can do anything. The challenge is building something
for which the final value exceeds the bill of materials.

Working with Brooks, Angle and Greiner realized that a whole
different approach was necessary for making robots a consumer
appliance as affordable and accessible as vacuum cleaners, micro-
waves, or washing machines. Drawing on Brooks’s ideas about
robots that combine simple behaviors to perform desired functions,
they removed nearly all the AI from the equation.
Consider how a typical AI researcher would design a robotic
vacuum cleaner. First, the robot would get a complete computer
vision system and software to maintain an internal map of each
room. The robot would calculate an optimum path to ensure that
it missed no spot of dirt on the floor, while identifying obstacles
in advance.
To do this, the robot would need cameras (probably with
stereoscopic vision) and software that could characterize and
locate objects in the room. It would also need some way to keep
track of which areas had been cleaned and which were still
dirty.
The iRobot team realized that robots with that level of intel-
ligence were complex and hard to make reliable. Indeed, even the
most advanced experimental AI robots would have problems deal-
ing with the changing environment of the home. What happens
when one of the objects in the robot’s internal map is the family
cat, which moves between the time of initial calculation and the
beginning of the cleaning pass? (It is doubtful the cat will appreci-
ate being vacuumed!)
Assuming they could even build such a robot, it would probably
cost about as much as a new car. Not many consumers will pay that
kind of price for a humble household appliance.

Behavioral Building Blocks

The iRobot team decided to take a radically different approach,
following Brooks’s principle of combining simple behaviors to get