Emergent adaptive behaviour of GRN-controlled simulated robots in a changing environment

Abstract

We developed a bio-inspired robot controller combining an artificial genome with an agent-based control system. The genome encodes a gene regulatory network (GRN) that is switched on by environmental cues and, following the rules of transcriptional regulation, provides output signals to actuators. Whereas the genome represents the full encoding of the transcriptional network, the agent-based system mimics the active regulatory network and signal transduction system also present in naturally occurring biological systems. Using such a design that separates the static from the conditionally active part of the gene regulatory network contributes to a better general adaptive behaviour. Here, we have explored the potential of our platform with respect to the evolution of adaptive behaviour, such as preying when food becomes scarce, in a complex and changing environment and show through simulations of swarm robots in an A-life environment that evolution of collective behaviour likely can be attributed to bio-inspired evolutionary processes acting at different levels, from the gene and the genome to the individual robot and robot population.

Using digital organisms to investigate the effect of whole genome duplication in (artificial) evolution

“…As supported by our simulations, and in previous research (24,25), in a stable environment, gradual evolution can successfully explain the divergence and optimization of evolutionary processes. When, on the other hand, the environment drastically changes, in a short geological time frame, gradual evolution has difficulty to keep up, while new species or life forms, if different enough from the original ones, can successfully occupy the 'new' niches that have become available (3).…”

“…With the purpose of investigating whether WGDs may confer an adaptive advantage in the face of drastic environmental changes or extinction events, such as abrupt reductions in available food resources, we have implemented here an improved version of our previously designed bio-inspired and agent-based computational framework (24,25). Two different environmental scenarios were set: food reduction was either introduced at a fixed time step during the simulation, or dynamically every time the total population reached 1000 individuals.…”

“…It should be noted that, at every time step and for every simulation, the population size of polyploids and non-polyploids was computed. Although, after food removal, a drop in population size was generally observed, when and how and to what extent the population size drops can greatly vary from simulation to simulation (25). It is inherent to the system that DOs with different genomic makeups and encoded GRNs can have very different responses to the same amount of food removal, which makes it hard to summarize the results, except for evaluating overall success when putting polyploids and non-polyploids in competition.…”

“…In the first scenario, food was dynamically removed every 60 time steps after the total population (polyploids plus non-polyploids) reached 1000 individuals. Starting from a population of 200 individuals at t0, having reached a population size of 1000 indicates that at least part of the population has adapted to the current environment and started to replicate rapidly (24,25). Under this first scenario, we performed 30 simulations under each of the following four food reduction levels: no food reduction, except for what is found and consumed by the DOs, 50% food reduction, 70% food reduction, and 85% food reduction.…”

“…actuators. The strengths and advantages of our bio-inspired agent-based modeling approach have already been tested under a relatively simple setup, such as emergent swarm behavior and overall enhanced adaptation in a changing environment (24,25).…”