A Comparative Analysis of Darwinian Asexual and Sexual Reproduction in Evolutionary Robotics

Evolutionary Robotics systems draw inspiration from natural evolution to solve the problem of robot design. A key moment in the evolutionary process is reproduction, when the genotype of one or more parents is inherited by their offspring. Existent approaches have used both sexual and asexual reproduction but a comparison between the two is still missing. In this work, we study the effects of sexual and asexual reproduction on the controllers of an Evolutionary Robotics system. In our system, both morphologies and controllers are jointly evolved to solve two separate tasks. We adopt the Triangle of Life framework, in which the controllers go through a phase of learning before reproduction. Using extensive simulations we show that sexual reproduction of the robots' brains is preferable over asexual reproduction as it obtains better robots in terms of fitness. Moreover, we show that sexually reproducing robots present different morphologies and behaviors than the asexually reproducing ones, even though the reproduction mechanism only affects their brains. Finally, we study the effects of the reproduction mechanism on the robots' learning capabilities. By measuring the difference between the inherited and the learned brain we find that robots that evolved using sexual reproduction have better inherited brains and are also better learners.