behavior based robotics
DESCRIPTION
BEHAVIOR BASED ROBOTICS. IN ABOUT 10 MINUTES. by Don McFarland. Classical Design Brooks’ Ideas Two BBR Examples Summary. Design by Introspection. I perceived a painful change I adjusted my internal model I planned my response I am acting out the plan. Classic Robot Architecture. - PowerPoint PPT PresentationTRANSCRIPT
BEHAVIOR BASED ROBOTICS
IN ABOUT 10 MINUTES
by Don McFarland
• Classical Design
• Brooks’ Ideas
• Two BBR Examples
• Summary
Design by Introspection
• I perceived a painful change
• I adjusted my internal model
• I planned my response
• I am acting out the plan
Classic Robot Architecture
FROM SENSORS TO ACTUATORS
• Stanford/Nilsson ‘69
• goal-directed agent
(STRIPS planning)• simple environment• slow as molasses
SHAKEY
The Classic Robot
A Different Approach
• 1984: Three groups: MIT(Agre,Chapman), SRI(Rosenschien, Kaelbling), MIT(Brooks)
• Tightly couple sensing and action
• Avoid internal (symbolic) world model
• Organize systems by behaviors
• A lot more …
ENVIRONMENTS
Reflex
Utility-Based
Goal-Based
Reflex + State
Simple Complex
Classical
BBR
Behavior-Based Robotics (BBR)
• “The world is its own model”
• Situated-ness: senses its surroundings
• Embodied: physically in the world
• Evolve systems from simple behaviors up to complex behaviors
Central Brooksian Ideas
An student-avoiding, pizza-seeking TA
confusionconf
usio
n
confusion
confusion
stomachnose
nose
nose
feet nose
PIZZA
CONFUSEDSTUDENT
Student avoider …
confusion avoid feet
… and pizza-seeker
stomach
feednose
confusion avoid feetS
FEED “SUBSUMES”
AVOID
“SUBSUMPTION ARCHITECTURE”
A Bug Example: Genghis
• Standup• Simple walk• Force balancing• Leg lifting• Whiskers• Pitch stabilization• Prowling• Steered prowling
Brooks’ Recent Work
COGKISMET
Cynthia Breazeal
Summary
• Brooks 1984 shifted from goal-directed agents in artificial/simulated worlds to reflex agents in dynamic, complex worlds.
• Idea 1: situated, embodied robots using the world as its own model.
• Idea 2: pattern after biological evolution (and design from simplest behaviors, overlaid by more complicated)
QUESTIONS?