Bio-mimetic Robot Control Companion slides for the book Bio-Inspired - PowerPoint PPT Presentation

Bio-mimetic Robot Control Companion slides for the book Bio-Inspired Artificial Intelligence: Theories, 1 Methods, and Technologies by Dario Floreano and Claudio Mattiussi, MIT Press

Mainstream A.I. • A.I. was born in 1956 as a research field to replicate human intelligence • Over the years, most efforts concentrated on reasoning, planning, logic Chess play was the hallmark of intelligence Until 1997: defeat of Kasparov by IBM DeepBlue Companion slides for the book Bio-Inspired Artificial Intelligence: Theories, 2 Methods, and Technologies by Dario Floreano and Claudio Mattiussi, MIT Press

Machina Speculatrix (Grey Walter, 1953) Simple analog electronics Simple phototaxis, obstacle avoidance Intelligence due to interaction with environment Reconstructed by Owen Holland Companion slides for the book Bio-Inspired Artificial Intelligence: Theories, 3 Methods, and Technologies by Dario Floreano and Claudio Mattiussi, MIT Press

Vehicles (Valentino Braitenberg, 1984) Show behavioral effect of simple patterns of neural connectivity Intelligence is in the eye of the observer Attraction Avoidance Curiosity Memory Learning Intentionality … Artificial Evolution Companion slides for the book Bio-Inspired Artificial Intelligence: Theories, 4 Methods, and Technologies by Dario Floreano and Claudio Mattiussi, MIT Press

What is Bio-mimetic Control? WHY Engineering motivation: Extract a principle from biology that solves a specific engineering problem that is not satisfactorily solved with other techniques Scientific motivation: Use a robot to validate a model that is heavily based on physical embodiment and environmental situatedness There is not a single methodology. Four case studies: - Vision-based flight (housefly) - Song recognition and localization (cricket) - Vision-based homing (ants, wasps) - Swim and walk (salamander) Companion slides for the book Bio-Inspired Artificial Intelligence: Theories, 5 Methods, and Technologies by Dario Floreano and Claudio Mattiussi, MIT Press

Vision-Based Flyers INDOOR FLIGHT Small size brings strong energy constraint, which rules out active sensors (active IR, sonar, laser) and computationally-expensive signal processing Companion slides for the book Bio-Inspired Artificial Intelligence: Theories, 6 Methods, and Technologies by Dario Floreano and Claudio Mattiussi, MIT Press

Insect Vision VISION-BASED INSECT FLIGHT Flies and several other insects use modifications of optical image ( Optic Flow ) to avoid obstacles. • Distance to objects is proportional to OF magnitude only in straight flight. • Houseflies fly along straight trajectories interrupted by rapid, saccade-like, rotations to avoid obstacle. Companion slides for the book Bio-Inspired Artificial Intelligence: Theories, 7 Methods, and Technologies by Dario Floreano and Claudio Mattiussi, MIT Press

Artificial Compound Eye VISION-BASED INSECT FLIGHT Franceschini, Pichon, Blanes, 1992 Circular array of analog Elementary Motion Detectors Analog electronics 1- Straight trajectory 2- Computation of obstacle distance map 3- New heading: goal direction and distance map Companion slides for the book Bio-Inspired Artificial Intelligence: Theories, 8 Methods, and Technologies by Dario Floreano and Claudio Mattiussi, MIT Press

Indoor Flying Robot Zufferey & Floreano, 2006 30 g Two linear cameras pointing 45% from front 15-minute autonomy - Front = small OF 1.2 - 2.5 m/s - Sides = large OF, but not useful for navigation Companion slides for the book Bio-Inspired Artificial Intelligence: Theories, 9 Methods, and Technologies by Dario Floreano and Claudio Mattiussi, MIT Press

When & Where to Turn Zufferey & Floreano, 2006 OFDiv = OFRight - OFLeft initiate rotation if OFDiv > threshold OFDiff = abs[OFRight] - abs[OFLeft] rotate towards OFDiff Companion slides for the book Bio-Inspired Artificial Intelligence: Theories, 10 Methods, and Technologies by Dario Floreano and Claudio Mattiussi, MIT Press

Roll Stabilization CONTROL STRATEGY • Halteres used to stabilize flight in housefly • MEMS piezo-electric gyroscopes in robot Companion slides for the book Bio-Inspired Artificial Intelligence: Theories, 11 Methods, and Technologies by Dario Floreano and Claudio Mattiussi, MIT Press

12

Cricket Song Recognition and Localization • Female cricket recognizes male song and goes towards it • Song is composed of syllables of pure tone at 4.5 kHz • Syllable Repetition Interval is different for each species • Female goes only to crickets of same species Companion slides for the book Bio-Inspired Artificial Intelligence: Theories, 13 Methods, and Technologies by Dario Floreano and Claudio Mattiussi, MIT Press

Existing Models OPEN ISSUES ON CRICKET’S BRAIN How does female recognize song? How does it go towards song source? Is firing rate or firing time used by auditory neurons? Several models exist, but none is complete and well-specified Companion slides for the book Bio-Inspired Artificial Intelligence: Theories, 14 Methods, and Technologies by Dario Floreano and Claudio Mattiussi, MIT Press

Integrated Model with Spiking Neurons ARTIFICIAL NEURONS Spiking neuron integration + leakage spike x1 refractory period x2 x3 Binary values x4 • Cricket advances at constant speed and turns when hears song • Synaptic strength decreased by controlateral AN, but exponential recovery • First AN to fire generates firing of ipsilateral MN • Too high Syllable Repetition Interval will neutralize both MN neurons • Too slow Syllable Repetition Interval will not generate correct trajectory Companion slides for the book Bio-Inspired Artificial Intelligence: Theories, 15 Methods, and Technologies by Dario Floreano and Claudio Mattiussi, MIT Press

Validation with Robot ROBOCRICKET, Edinburgh Analog electronics model cricket auditory system Robot can track recorded songs from real crickets Robot ignores songs of other species Companion slides for the book Bio-Inspired Artificial Intelligence: Theories, 16 Methods, and Technologies by Dario Floreano and Claudio Mattiussi, MIT Press

Desert Ant Homing DESERT ANT NAVIGATION No pheromone use because of evaporation + scattered PREY food Strategies : • Path integration • Visual piloting • Systematic search 17 NEST

Snapshot Matching Average Landmark Vector model Insect may store retinotopic images of (Lambrinos et al, 2000): the visual scene and recover paths by One vector per landmark aligning themselves so to match stored ALV: average of all landmark vectors templates (Cartwright & Collett, 1983): Storage of target ALV • Assumes high memory capacity Heading = ALV(cur) - ALV(tar) • Neural circuit may have to be complex Companion slides for the book Bio-Inspired Artificial Intelligence: Theories, 18 Methods, and Technologies by Dario Floreano and Claudio Mattiussi, MIT Press

Sahabot VISUAL SNAPSHOTS Companion slides for the book Bio-Inspired Artificial Intelligence: Theories, 19 Methods, and Technologies by Dario Floreano and Claudio Mattiussi, MIT Press

Homing by Image Navigation Zeil et al. 2003 Root Mean Square error of images form a surface with an increasing gradient as one moves away from home (target image) Insects may memorize home image at outward flight and compute RMS to sample gradient on return flight Companion slides for the book Bio-Inspired Artificial Intelligence: Theories, 20 Methods, and Technologies by Dario Floreano and Claudio Mattiussi, MIT Press

Rythmic Locomotion • CPG is a set of neurons, that display rythmic oscillatory activity • When coupled with body, can be used to generate rythmic motion • Architecture, weight strengths, and time delays of connection affect behavior Amphibot, Ijspeert EPFL CPG neurons oscillate within a frequency range Below or above that range, they stop working Companion slides for the book Bio-Inspired Artificial Intelligence: Theories, 21 Methods, and Technologies by Dario Floreano and Claudio Mattiussi, MIT Press

Salamander’s behavior • Salamanders can both swim and walk in wave motion • Speed of waves are proportional to intensity of brain signals to CPG • Transition between walk and swim brain signal - behavior high fast swim Open questions: swim • One, two, or more CPG circuits? transition • Architecture? fast walk walk low still Companion slides for the book Bio-Inspired Artificial Intelligence: Theories, 22 Methods, and Technologies by Dario Floreano and Claudio Mattiussi, MIT Press

Model of Salamander’s control circuit MODEL ASSUMPTIONS (Ijspeert et al. 2007): • Body CPG + Limb CPG • Limb CPG oscillates at lower frequency and saturates at lower threshold • Connections from limb to body CPG are stronger than connections within body CPG Companion slides for the book Bio-Inspired Artificial Intelligence: Theories, 23 Methods, and Technologies by Dario Floreano and Claudio Mattiussi, MIT Press

Transition between walk and swim Companion slides for the book Bio-Inspired Artificial Intelligence: Theories, 24 Methods, and Technologies by Dario Floreano and Claudio Mattiussi, MIT Press