Using robots to model animals: a cricket test Barbara Webb

1
Using robots to model animals a cricket
testBarbara Webb

• Presenter Gholamreza Haffari

2
Bio-Robotics

• A methodology established to bridge the gap
  between Artificial Intelligence Biology
  (Kortmann 1998)
• Based on the method of explanation by modeling

3
Related fields

• Neuroethology
• Subfield of Biology
• How animal behavior is rooted in the neural
  systems in the animal brain
• Computational Neuroethology
• Studying neural mechanisms that underlie adaptive
  behavior by building autonomous agents
• Not restricted to modeling existing animals
  (natural as well as artificial agents)

4
Related fields (cont.)

• Animat
• Studying natural adaptive systems by building
  artificial autonomous agents
• Belongs to behavior-based approach to AI

5
Bio-Robotics Methodology

• Models are built, based on the hypotheses on
  neural mechanisms that underlie adaptive behavior
  in real animals
• By careful experimentation with the model and
  reliable interpretation of its behavior, one can
  obtain evidence on the hypothesized mechanism
  modeled

6
Framework of Modeling
A general framework of modeling (Kortmann 1998)
7
Modeling animals

• It is necessary to accurately represent the real
  physical interaction of the animal and the
  environment
• Physical environment and physical interactions
  are extremely complex to model symbolically
• Keeping in mind that biological systems are
• Situated in the environment and interact with it
• Embodied, i.e. they always have a body
• Robots are well suited to being used as physical
  models of animals

8
Studying simple animals

• It is relatively easy to find isolated adaptive
  behaviors that are likely to be pre-programmed by
  a simple direct pathway in their brain
• These pathways are expected to be found
  relatively easily

9
Phonotaxis behavior of cricket

• Ability of the female cricket to find a
  conspecific male by walking or flying towards the
  calling song the male produces
• Conspecific means (individual) from the same
  species
• Getting to the target by constantly adjusting the
  direction according to current sensory cues

10
Directionality and Recognition

• How does the cricket identify the correct signal?
• How does it detect the difference between two
  sides, and hence choose which way to turn?

11
Mechanisms underlying phonotaxis

• Until now, precise sensory motor control of the
  phonotaxis behavior has not been found
• Webb (1993) decided to design a robot model to
  give evidence for a hypothesized control mechanism

12
Phonotaxis mechanisms (cont.)

• The phonotaxis mechanism can be divided into
• two components
• The peripheral auditory system (the ears)
• Its working can be described in the physical
  level
• The brain mechanism
• Which is described in the language of
  neurophysiology

13
Peripheral auditory system

• Consists of
• Two auditory organs located in the forelegs
• An H-shaped tracheal tube that leads through the
  body and have four ends
• Two in the forelegs (the tympani)
• Two at the side of the thorax (the spiracles)

14
Peripheral auditory organ
(Kortmann 1998)
15
Physics of the auditory system

• In a simplified version, consider only the
  connection between the two tympani
• Two external and internal sound waves reach the
  tympanum
• The external signal comes directly from the
  source
• The internal signal comes indirectly from
  contralateral tympanum via tracheal tube

16
Physics of auditory sys. (Cont)

• Assume the length between the two tympani to be ¼
  of the wavelength of the calling song
• Sound waves arrive to the closest tympanum in
  antiphase from opposite sides, and cause the
  optimal response of the membrane
• But, sound waves arrive in phase to the other
  tympanum and cause the minimal response

17
Phase cancellation
18
Robots auditory mechanism

• Two miniature microphones positioned 4.5 cm apart
  from each other (1/4 the wavelength of the 2 kHz
  signal used)
• Input at the left ear is combined with the
  delayed signal from the right ear
• The same occurs for the right ear

19
Comparing the response

• Signals of auditory receptors are carried by
  auditory nerves to small number of interneurons,
  one pair of these (AN1) appears particulary
  receptive
• The comparison can be based on the
• Firing rate
• Latency

20
Robots comparison mechanism

• Response-dependent latency is implemented in the
  robot by using summation with decay
• Consider variables anR and anLfor each ear, where
  for each variable
• Each an fires when it is gt 8 (behaves like a
  low-pass filter)
• The comparison then occurs in the module COMPARE
  (based on the onset of an variables), increasing
  the value of turning-tendencyLor turning-tendencyR

21
Terminology in describing songs
(Kortmann 1998)
22
Experiments setup

• Frequency 2 kHz
• Syllable repetition rate 1.6 Hz
• Speaker and Robots
• starts positions

23
Locating the sound source
24
Locating the sound source with obstacles
25
Recognizing the sound source

• Is the behavior selective, i.e. does the robot
  approach no-ideal sound sources?
• Slow syllable rate (1 Hz) Fast Syllable rate
  (2.5 Hz)

26
Effect of chirps

• The syllables are repeated only a few times and
  these groups (chirps) are separated by equal
  length of silence
• Without chirps, there is a certain amount of
  vacillation in the approach to the sound
• By chirps, cricket makes only occasional
  adjustment of heading rather than continual
  adjustment

27
Effect of chirps (cont.)
Three-syllable chirps at the rate of 3 Hz
28
Choice phenomenon

• Female cricket seems able to choose to approach
  directly just one singing male despite a number
  of other males also singing well within auditory
  range
• Does it imply central complex processing ?
• No, it can be seen in the behavior our simple
  robot

29
Choice phenomenon (cont.)
30
Discussion

• Understanding biological systems provide a set of
  tricks that may usefully be adapted for
  robotics
• Furthermore, they lead to better explanation of
  the behavior
• The mechanisms in biological systems are the
  result of the evolution and thus may rarely
  represent ideal methods for achieving the task
  but they are good enough

31
Using simulation model?

• Even in the most exhaustive simulations some
  potentially important effects may be neglected,
  overlooked or improperly modeled
• It is often not reasonable to attempt to
  account for the complexity and unpredictability
  of the real world

32
Conclusion

• The links between biology and robotics have
  tended to be at an abstract level
• At the level of behavior
• Not representing sensory transduction, neural
  processing and motor control
• Detailed attention to one highly specific animal
  competence will contribute to a general
  understanding of the functioning of sensorimotor
  mechanisms

33
Thanks