Biologically Inspired Embodied Evolution of Survival

Abstract

Embodied evolution is a methodology for evolutionary robotics that mimics the distributed, asynchronous and autonomous properties of biological evolution. The evaluation, selection and reproduction are carried out by and between the robots, without any need for human intervention. In this paper we propose a biologically inspired embodied evolution framework, which fully integrates self-preservation, recharging from external batteries in the environment, and self-reproduction, pair-wise exchange of genetic material, into a survival system. The individuals are, explicitly, evaluated for the performance of the battery capturing task, but also, implicitly, for the mating task by the fact that an individual that mates frequently has larger probability to spread its gene in the population. We have evaluated our method in simulation experiments and the simulation results show that the solutions obtained by our embodied evolution method were able to optimize the two survival tasks, battery capturing and mating, simultaneously. We have also performed preliminary experiments in hardware, with promising results.

What is situated evolution?

“…One such example is that of Elfwing et al [9], [10]. There, the robots have to harvest batteries to update their energy level.…”

“…Donor selection Host selection Egg selection Embodied evolution [11] agents push own genetic material at at rate proportional to their fitness (with known maximum) agents resist 'infection' proportionally to their fitness implicit: an agent updates its own genetic material/controller [30] equivalent to [11] [9] unclear when and how it takes place (is it among the virtual agents in a robot, for instance), but it is a separate step combined into a single replacement step when a virtual agent dies (presumably after a fixed amount of time) [26] push model: individual to send (migrate) is selected through roulette wheel, then broadcast (presumably) with a probablility based on its relative fitness fitness value from the sender is ignored either the robot itself or -if seen at the level of the local GA's individuals-tournament with the worst locally available individual [19] equivalent to [11] [24] synchronised by internal timers; the robots emit a "mating call" over the radio, where they 'shout' their identification, fitness values, and chromosomes best controllers mate (unclear how they are coupled) and survive, the worst robots are reprogrammed with the results; exact nature of selection is unclear [27] implicit (as always for evolutionary strategies)…”

Open-ended on-board Evolutionary Robotics for robot swarms

“…Most follow up -if sometimes only notionally-on Embodied Evolution, which was introduced by Watson, Ficici and Pollack [22], [49] and relies on broadcast of (mutated) genes at a rate proportional to the robot fitness. Some extensions of these original works were conducted either by introducing a maturation period [51], or by implementing time-sharing to cope with small populations [19], [18], [48]. Other work focuses on the competitive diffusion of genotype through comparing fitness value [43], spatially structured evolution strategies [25] and other local deterministic or tournament selection scheme using various genotypes and fitness values diffusion algorithms [11], [17], [3], [50].…”

Is situated evolution an alternative for classical evolution?

“…From the 5 input sensors, 1 is for detection of hits against obstacles (the variation between the estimated trajectory and the real trajectory when hitting against an obstacle, see Figure 5) and 4 for the detection of the number of agents within each of the four sectors of vision (see Figure 4). Vision: figure 4 shows the vision range to detect (and cover) cells, which is an ellipse with one focus in the position of the watcher and another focus on the major axis following the orientation angle 2 . The distance between these foci is 40 pixels, and the sum of distances from every point within the vision range to both foci is less than 60 pixels.…”

“…These implementations employ time-sharing to evaluate an on-board pool of individuals [10], [2], [7]. Individuals migrate from one robot to another using local communication.…”