The salamander presents the interesting property of being able of both
aquatic and terrestrial locomotion. Analyzing its locomotor circuitry
may therefore give some insights into the changes of locomotor
circuits which have accompanied this transition during vertebrate
evolution. In order to investigate salamander locomotion and
visually-guided behavior, we developed a 3D biomechanical simulation
of the salamander's body whose muscle contraction is determined by the
locomotion controller simulated as a leaky-integrator neural network
[1, 2]. While the connectivity of the neural circuit underlying
locomotion in the salamander has not been decoded for the moment, this
work presents the design of a neural circuit which has a general
organization corresponding to that hypothesized by neurobiologists [3,
4]. The locomotion controller is based on a body central pattern
generator (CPG) corresponding to a lamprey-like swimming controller,
and is extended by a limb CPG for controlling the salamander's limb
(Figure 1). The complete locomotor circuit is developed in three
stages with first the development of segmental oscillators, second the
development of intersegmental coupling for the making of a
lamprey-like swimming CPG, and finally the development of the limb CPG
and its coupling with the body CPG [1, 5]. An evolutionary algorithm
is used to instantiate all the parameters of the neural circuits
(synaptic weights and time parameters) for the different stages given
a high level description of the desired state space trajectories of
the different subnetworks. A similar approach had previously been used
to successfully generate swimming controllers for a model of the
lamprey [6, 7]. Here the fitness functions are designed to
reward neural networks which can produce stable oscillations whose
frequency can be modulated by the level of an input signal (tonic
drive), and which, when coupled together, optimize the speed of locomotion
of the mechanical simulation. A controller is thus developed which
can produce a neural activity and locomotion gaits very similar to
those observed in the real salamander [4, 8], that is, an anguiliform
swimming gait in water and a trotting gait on ground in which the body
makes an S-shaped standing wave coordinated with the movements of the
limbs (see http://rana.usc.edu:8376/~ijspeert/salamander.html
for several animated gifs). By simply varying the tonic excitation
applied to the network, the speed, direction and type of gait can be
varied.
We are currently developing a model of the salamander's visual system,
with the purpose 1) to investigate the neural mechanisms underlying
the transformation of visual inputs into motor commands, and 2) to
investigate the dynamics which results from connecting a visual system
to a locomotor system. Based on comparative studies of frog and
salamander visual systems [10, 11, 12], models of the salamander's
retinas, optic tectum, and pretectum are therefore integrated into the
biomechanical model of the salamander and connected to the locomotor
circuit via a model of the brainstem. We are in particular interested
in analyzing the effects of a moving body on visual perception, and
the robustness of the pattern generation of the locomotor circuitry
against constantly varying commands. Preliminary experiments [9] show
that the simulated salamander with the locomotor circuit connected to
a simplified visual system is capable of robustly tracking a randomly
moving target (Figure 2).
References:
[1] Ijspeert, A., Hallam, J., and Willshaw, D. (1998). From lampreys to salamanders: evolving neural controllers for swimming and walking. In Pfeifer, R., Blumberg, B., Meyer, J.-A., and Wilson, S., (Eds.), From Animals to Animats, Proceedings of the Fifth International Conference of The Society for Adaptive Behavior (SAB98), pages 390-399. MIT Press.
[2] Ijspeert, A. (2000). A 3-D biomechanical model of the salamander. Proceedings of the Second International Conference on Virtual Worlds, Paris, France, 5-7 July 2000. Springer-Verlag. (Accepted for publication).
[3] Cohen, A. (1988). Evolution of the vertebrate central pattern generator for locomotion. In Cohen, A. H., Rossignol, S., and Grillner, S., (Eds.), Neural control of rhythmic movements in vertebrates. Jon Wiley & Sons.
[4] Delvolve, I., Bem, T., and Cabelguen, J.-M. (1997). Epaxial and limb muscle activity during swimming and terrestrial stepping in the adult newt, pleurodeles waltl. Journal of Neurophysiology, 78:638-650.
[5] Ijspeert, A. (2000). A connectionist central pattern generator for the swimming and trotting of a simulated salamander. Submitted to Biological Cybernetics.
[6] Ijspeert, A., Hallam, J., and Willshaw, D. (1999). Evolving swimming controllers for a simulated lamprey with inspiration from neurobiology. Adaptive Behavior, 7(2). 151-172.
[7] Ijspeert, A., and Kodjabachian. (1999). Evolution and development of a central pattern generator for the swimming of a lamprey, Artificial Life 5:3, pp 247-269.
[8] Frolich, L. and Biewener, A. (1992). Kinematic and electromyographic analysis of the functional role of the body axis during terrestrial and aquatic locomotion in the salamander ambystoma tigrinum. Journal of Experimental Biology, 62:107-130.
[9] A.J. Ijspeert, M. Arbib (2000). Visual tracking in simulated salamander locomotion, From Animals to Animats, Proceedings of the 6th International Conference on the Simulation of Adaptive Behavior (SAB2000), Paris, September 11-15. (Accepted for publication).
[10] Arbib, M. (1987). Levels of modeling of visually guided behavior. Behavioral and Brain Sciences, 10:407-465.
[11] Arbib, M. and Liaw, J. (1995). Sensorimotor transformations in the world of frogs and robots. Artificial Intelligence, 72:53-79.
[12] Lamb, M. (1997). Modeling behavior-based depth vision in frog and salamander. PhD thesis, Department of Neuroscience, University of Southern California.
Preprints of some of these articles can be downloaded from http://rana.usc.edu:8376/~ijspeert/publications.html.