self-organization, embodiment, and biologically inspired robotics rolf pfeifer, max lungarella,...
TRANSCRIPT
Self-Organization, Embodiment, and Biologically Inspired Robotics
Rolf Pfeifer, Max Lungarella, Fumiya IidaScience – Nov 2007.
Rakesh GosangiPRISM lab
Department of Computer Science and Engineering
Texas A&M University
Outline
• Embodiment• Sensory-Motor Coordination• Embodiment and Information• Passive Dynamics• Designing Morphology• Future
Embodiment
• Embodied cognition– Human cognition is shaped by
• Human morphology• Interaction with the environment
• Example– Motor theory of speech perception
• Embodied cognition contradicts computational theory of mind
Computational Mind
Embodied cognition
Classical Robotics
• Centralized control– Microprocessor– Information processing system
• Disadvantages– Energetically inefficient– Cannot learn from interaction– Lack adaptivity
Embodied Systems
• Distribute the control– Controller– Morphology
• Self-organization
– Materials• Functional materials• Simplify neural control
Outline
• Embodiment• Sensory-Motor Coordination• Embodiment and Information• Passive Dynamics• Designing Morphology• Future
Sensory-Motor coordination
• Interactions where sensory stimulation influences motor actions and motor actions in turn influence the sensory stimulation.
• Example – looking at an object in hand (foveation)
• Dependence between sensor,neural, and motor variables
Movement
Sensory stimulation
Induces
Influences
Properties of sensory-motor systems
• Sensory and motor processes are coupled– Neither one is primary
• Correlation between different sensory modalities• Temporal and spatial patterns in correlation
– Characterize robot-environment interaction
Examples
• Salamander robot– Switching between swimming and walking (video)
• Visual homing– How bees and wasps find their way back home
• Phonotaxis– How female crickets identify male crickets in noisy
environment
Outline
• Embodiment• Sensory-Motor Coordination• Embodiment and Information• Passive Dynamics• Designing Morphology• Future
Information theoretic implications
• Redundancy across sensory channels• Information structure develops with interaction
with environment• Changes in morphology effect the information
structure• Learning effects information structure• Learn cross-modal associations
– Correlations between different sensory modalities
Information self structuring
• A – foveation– camera tracks the ball
• B – random– Camera movement is unrelated
to the ballMeasures are applied to the camera image
Images adapted from M. Lungarella, O. Sporns, PLoS Comp. Biol. 2, e144 (2006)
Outline
• Embodiment• Sensory-Motor Coordination• Embodiment and Information• Passive Dynamics• Designing Morphology• Future
Passive Dynamics
• “Intelligence by mechanics”• Intrinsic dynamics of the mechanical system
yields self-stabilizing behavior• Select morphology and materials to exploit
physical constraints in ecological niche• Examples
Passive walking
• Walking down a slope– Without control or actuation– Self stabilizing using gravity– Passive Walking
• Walking on flat surfaces– Active power source to replace gravity– Reinforcement learning to find a policy that stabilizes
the robot– Use less energy and control compared to powered
robots– Passive Dynamic Walking
Other examples
• Ornithopters– Passive dynamics for wing rotation– Video
• Waalbot– Adhesive materials like gecko– Video
Outline
• Embodiment• Sensory-Motor Coordination• Embodiment and Information• Passive Dynamics• Designing Morphology• Future
Morphology
• Evolutionary optimization of robot morphology– Current work on evolving the robot controller
• “Morphofunctional” machines– Change functionality by modifying morphology
• Increase adaptability, versatility and resilience
Self reconfigurable robots
• Macroscopic modules• Size of the modules constraint the morphology
and functionality• Magnetic or mechanic docking interfaces
• Self assembling modular robot• Simulation Video
Outline
• Embodiment• Sensory-Motor Coordination• Embodiment and Information• Passive Dynamics• Designing Morphology• Future
Future
• Imitation learning– Learn from humans and other robots reducing the
search space
• Collective robotics– With material and morphological considerations
• Self-replicating robots– Machines that can autonomously construct a functional
copy of themselves– John Von Neumann