Learning in a virtual model of the C. elegans nematode for locomotion and chemotaxis

In this paper, we propose a learning control system, which models the neural circuits controlling locomotion and chemotaxis of the Caenorhabditis elegans nematode. Using the realistic 3D-simulator of the nematode, we have conducted a series of successful experiments in teaching the proposed model. It is shown that the control system can stably learn an effective way of movement forward in 100 working cycles on the average, and identify an optimal chemotaxis strategy in 1000 cycles on the average. At the same time, we observe a considerable visual likeness between the behavior of the model and the behavior of a real nematode and noted the coincidence of the detected chemotaxis strategy with the strategy used by the biological prototype. The results of experiments have shown that the movement function and associated orientation mechanisms of a nematode can be obtained by way of teaching only in interaction with the environment, and the proposed model of control system is quite effective and can be successfully used to control complex objects with many degrees of freedom.