Training of a “smart” triangular swimmer with the help of genetic algorithms

Abstract

Various microorganisms in nature use diverse non-reciprocal swimming gaits in search of nutrition or prey or to escape predators. A very common strategy comprises non-reciprocal deformation of the shape, ensuring propulsion in the medium. Inspired by natural swimmers, efforts have been made to design artificial swimmers to perform specific tasks, such as targeted drug delivery in the case of nanomedical applications. In this work, we train a two-dimensional, triangular swimmer [1] to move in a desired direction using different propulsion gaits (see Fig.1). This training is achieved with adaptive neural networks (which connect the degrees of freedom and the forces acting on the swimmer), involving thereby the NEAT algorithm [2] which optimizes the internal architecture of these networks. While in the previous work a simple one-dimensional, linear three-bead swimmer was trained to swim in a chemical landscape [3], here we focus on the considerably more challenging case of the triangular swimmer (see Fig. 1(a)-(c)): flapping, chiral and walking respectively are three emergent non-reciprocal propulsion modes. Since in two dimensions, a simple reward, for example, displacement, cannot induce motility in the swimmer, and in the absence of a well-defined reward scheme, the swimmer either stays stationary or rotates in a circular path with no net displacement, it is important to introduce a complex reward scheme which is a function of instantaneous displacement, average displacement, angular displacement, shape factor. We also demonstrate that in this setup, the swimmer can be trained to propagate in a desired direction and to develop swimming gaits that allows to find nutrient in a chemical landscape. Similar as in Ref. [1], we are able to extract valuable information about the swimmer’s strategies by analyzing the internal structure of the emerging networks. Notably, the former emergent networks are simple in nature though visibly different for different swimming gaits.