Table 2.

Glossary.

Field	Comment
Implementation	Distributed implementations have one genome for each robot, and an offspring is created only as the result of a mating event or by mutating the current genome. Hybrid implementations have multiple genomes per robot, and offspring can be created from this internal pool and from genomes “imported” through mating events. As stated earlier, the encapsulated scheme is not considered embodied evolution as there is no exchange of genomes between robots in this case.
	The experiments can use real robots or simulation.
Robot behavior	A monomorphic population contains individuals with similar genotypes (with variations due to mutation). A polymorphic population is divided into two (or more) subgroups of genetically similar individuals, and different genotypic signatures from one group to the other, e.g., to achieve specialization.
	We distinguish between experiments that target efficient individual behavior vs. collective behaviors (i.e., social behaviors, incl. cooperation)
Experimental settings	Identifies the task(s) considered in the experiment, e.g., obstacle avoidance, foraging, … None indicates that there is no user-defined task and that consequently, selection pressure results from the environment only. The number of robots used is also included. n1 − n2 indicates the interval for one experiment and n1; n2 gives numbers for two experiments.
Mating conditions	Mating can be based on proximity: two robots can mate whenever they are physically close to each other (e.g., in infrared communication range). In panmictic systems, robots can mate with all other robots, regardless of their location. Other comprises systems where robots maintain an explicit list of potential mates (a social network), which may be maintained through gossiping.
Selection scheme	Parents are selected from the received and internal genomes on the basis of their performance if a task is defined. Random parent selection implies only environment-driven selection. Currently, the only examples of other selection schemes use genotypic distance, but this category also covers metrics such as novelty.
Replacement scheme	Genomes can have a fixed lifetime, variable lifetime, or limited lifetime (similar to variable lifetime, but with an upper bound). Event-based replacement schemes do not depend on time but on events such as reception of genetic material (e.g., in the microbial GA used by Watson et al. (2002)).