Presented at Main Meeting and Dynamical Neuroscience Symposium.
Neural networks for chemotaxis in C. elegans:
rule extraction and robotics
T. C. Ferrée, T. M. Morse and S. R. Lockery
Institute of Neuroscience, University of Oregon, Eugene, OR 97403.
C. elegans moves up chemical gradients by detecting changes in concentration at the tip of its nose, and biasing its movement toward higher concentrations. The neural circuit for chemotaxis includes 40 neurons with both feedforward and feedback connections (C. Bargmann, unpublished). Electrical recordings from neurons in the anterior nerve ring of C. elegans suggest that they are effectively isopotential, and do not fire action potentials. We have shown previously that nonlinear networks of this type can be optimized to control chemotaxis in computer simulations of the nematode. To begin to understand how these networks function, we focus here on linear recurrent networks, which also produce effective chemotaxis control. We first derive an analytic solution for the body rate of turning in response to sensory input at the nose, then expand this solution in time derivatives of the sensory input. From this we extract simple, intuitive rules for chemotaxis which elucidate computations being performed implicitly by the network. Network models of this type can also be used to control autonomous vehicles. We use this approach to control phototaxis in a wheeled robot moving in two dimensions.
This work is supported by NIMH MH11373, NIMH MH51383, NSF IBN 9458102, ONR N00014-94-1-0642, the Sloan Foundation, and the Searle Scholars Program.