Real Robots Don’t Drive Straight

Fred MartinUMass Lowell

PEDAGOGICAL GOALSOF USING ROBOTS

• fun & engagement• learning engineering• learning AI

let’s look at parallel challengesfor engineering and AI educators

FEEDBACK: THE CENTRALENGINEERING CONCEPT

• Feedback is pervasive. Weare doing it all the time…

FEEDBACK: THE CENTRALENGINEERING CONCEPT

• Feedback is pervasive. Weare doing it all the time…

• standing

FEEDBACK: THE CENTRALENGINEERING CONCEPT

• Feedback is pervasive. Weare doing it all the time…

• standing• walking

FEEDBACK: THE CENTRALENGINEERING CONCEPT

• Feedback is pervasive. Weare doing it all the time…

• standing• walking• driving

FEEDBACK: THE CENTRALENGINEERING CONCEPT

• Feedback is pervasive. Weare doing it all the time…

• standing• walking• driving• maintaining conversationaldistance

FEEDBACK: THE CENTRALENGINEERING CONCEPT

• Feedback is pervasive. Weare doing it all the time…

• standing• walking• driving• maintaining conversationaldistance

don’t forget autonomic processes, likemaintaining body temperature, etc…

FEEDBACK IS INVISIBLE

We literally are not aware that weare doing it

Except when we are learningsomething new (or when perceptionbreaks down)

• steering a bicycle• swimming in a lane

STUDENTS DON’T BELIEVE INFEEDBACK

• “omnicient robot fallacy”• imperative programming• “drunkard’s walk” is anoutlandish idea

• “why doesn’t it drivestraight?!”

IS A ROBOT A MACHINE OR ACREATURE?

it depends on how we frame thetask…

CASE-WESTERN EGG HUNT CONTEST

CASE-WESTERN EGG HUNT CONTEST

FIRST LEGO LEAGUE 2004 CONTEST

FIRST LEGO LEAGUE 2004 CONTEST

THE SOLUTION: MAKE DRIVINGSTRAIGHT A PRIMITIVE!

• add quad encdrto motor

• default motioncmd is inrevolutions,not time

• can keytogether left &right motors

CURE WORSE THAN DISEASE

• much evidence that middle-schoolers understandconditionals and feedbackwhen given good materials

• community learns over time• these are kids’ first introto robotics

MOVING ON TO AI

• AI educators want to teach AI• students have too much funbuilding

• students spend too much timedebugging HW

• but…

AI & CLASSROOM ROBOTS:NOT SO MUCH?

• classical, knowledge-based AIassumes a perfectly representedworld (e.g. search, expertsystems, planning, game-playing)

• forcing it on sloppy classroom’bots makes the AI look bad

• students might have fun & learnthe AI, but do they believe init?

NEWER AI ON ROBOTS: YES

• Greenwald and Artz’s neural andBayesian networks to process IRreflectance sensors (and extractsignal from lots of noise)

• Thrun’s probabilistic robotics• and of course, Brooks’ originalprovocation of reactive robots

REAL ROBOTSDON’T DRIVE STRAIGHT

• things that look straightaren’t

• feedback is a powerful idea• kids are smart -- let’s nothold back the good stuff

• be creative to find AI thatworks well on sloppy systems