*Zoom*[PhD Thesis Presentation] - Ms. Farzana Rahman "Identifying the Evolutionary Conditions for the Emergence of Alternative Reproductive Strategies in Simulated Robot Colonies"


2020年4月30日 (木) 16:00 17:00


C210, Level C Center Bld.


Presenter: Ms. Farzana Rahman

Supervisor: Prof. Kenji Doya

Unit: Neural Computation Unit

Title: Identifying the Evolutionary Conditions for the Emergence of Alternative Reproductive Strategies in Simulated Robot Colonies

Webinar URL: no more available


Background and Aim
In many animals, individuals within one sex evolve with different behavior in order to increase their reproductive fitness. These evolved reproductive behaviors are called Alternative Reproductive Tactics (ARTs). Existing theoretical frameworks of Alternative reproductive tactics (mainly modeled by abstract mathematical models or individual based simulation) usually assume that there already exist two or three distinct phenotypes (with different tactic) and make prediction about the expected frequencies of different tactics in different situation. Also in these models agents have low embodiment with environment and lose important aspects of real life features between agents, such as- occupied body space of the agents, sensory information about the environment, physical interaction with other agents etc. By contrast, how distinct types emerge from initial continuous characteristics through evolution is a conceptually different question, which has been ignored by theoretical studies of ARTs, as they usually focus on how already present variations of reproductive behavior are maintained.
In this thesis, I developed a simulated robot evolution framework where agents have high embodiment with environment, along with both intersexual and intrasexual interaction, and investigated under which evolutionary condition alternative reproductive behavior spontaneously emerge from initially monomorphic population.

Materials and Methods
First, I designed and developed a robot simulator incorporating biological features (size, growth, mortality, sex, reproductive cost, etc.) with physical interaction, and a two layer robot controller, where the upper layer is a neural network that makes decision from the environmental input about choosing the behavioral module and the lower layer is four behavioral modules (Foraging, Mating or avoiding, Pushing or following, and waiting). Then I run evolution in this robot simulator where robotic agents can have intersexual and intrasexual interaction and thus evolve through both natural selection and sexual selection from initially monomorphic population.
After ensuring that the agent populations survive and evolve, I investigate the male and female agent behavior to detect any alternative reproductive tactics from the evolved population.
Second, I consider important environmental parameter (Food level, Mating cost, Competition level) and investigate for each parameter what kind of reproductive tactics evolve for each sex and how they are maintained for different level of each parameter. I investigate each parameter and also combination of them.
Then I investigate the evolutionary stability of the evolved phenotype of each sex and demonstrate how evolutionary stable point varies with the change of each environmental parameter.

I have so far obtained the following results.
In the evolved generation, two types of genetically distinct phenotypes with different tactics emerged in males- Bourgeois and sneakers, where they compete to get access to female. The Bourgeois males have high growth rate (that incurs high growth cost), and invest more of their time to look for food to maintain big body size, so that they can get less interruption during mating (win the competition to get access to the female) and increase their fitness (chance of getting offspring). The Sneaker males have low growth rate (low growth cost) and invest most of their time looking for female. As they cannot win competition (get access to female) with the Bourgeois male they flock around the Bourgeois male and when a female comes near they sneak choosing the mating module. They increase their fitness by following and mating with female when bourgeois male invest time to look for food and sneaking female when they come to mate with bourgeois male. But mean fitness (number of offspring per male) of sneaker male is about the same or lower than bourgeois male as bourgeois male is more favored by female and less interrupted during mating.
Females were also observed to evolve two types of genetically distinct phenotype with different tactics, where they mainly differ by offspring number and quality. They are- Quality oriented female (QOF) and Number oriented female (NOF). QOF were big in size, less interested in mating and produced small number of high quality (high size and energy) offsprings, where NOF were small in size, highly interested in mating and produced high number of low quality (low size and energy) offspring.

Further, I observed how different population evolved in different Food level (different number of food resource) and competition Level (male female ratio in the population). I found that Female ART is regulating with food level and Male ART with competition level.

The findings for the variation of the parameters (Food level, competition level and mating cost) and evolutionary stable state of the evolved phenotype will be further discussed by the time of thesis submission.

All-OIST Category: 

Subscribe to the OIST Calendar: Right-click to download, then open in your calendar application.