This paper investigates the evolution of modular robots using different selection preferences (i.e., fitness functions), aiming at novelty, speed of locomotion, number of limbs, and combinations of these. The outcomes are analyzed from different perspectives: sampling of the search space, evolved morphologies, and evolved behaviors. This results in a wealth of findings, including a surprise about the number of sampled regions of the search space and the effect of different fitness functions on the evolved morphologies.

Embodied evolution aims at self-sufficient adaptation in robot collectives in their task environment. An open question is how to achieve a good balance of effort over multiple tasks using embodied evolution. Most efforts to date rely on switching between predefined behaviours or on spatial or temporal separation of the tasks to achieve this. The research presented here is part of an effort to enable embodied evolutionary systems to achieve a balanced distribution of effort over multiple tasks without predefined behaviour and without any separation of the tasks. We propose and experimentally evaluate a selection mechanism that introduces a local reproductive advantage to individuals that specialise in underrepresented tasks. The paper shows that an embodied evolution implementation with this mechanism leads to balanced populations of generalist individuals, even when the environment severely penalises generalist behaviour. An extension that combines task-based and purely environmental selection in some cases leads to a balanced population of specialists, but does so inconsistently.

Evolving robot morphologies implies the need for lifetime learning so that newborn robots can learn to manipulate their bodies. An individual’s morphology will obviously combine traits of all its parents; it must adapt its own controller to suit its morphology, and cannot rely on the controller of any one parent to perform well without adaptation. This paper investigates the practicability and benefits of Lamarckian evolution in this setting. Implementing lifetime learning by means of on-line evolution, we first establish the suitability of an indirect encoding scheme that combines Compositional Pattern Producing Networks (CPPNs) and Central Pattern Generators (CPGs) as a relevant learner and controller for open-loop gait controllers. We then analyze a Lamarckian set-up and the effect of the parental genetic material on the early convergence to good locomotion performance.

Social learning enables multiple robots to share learned experiences while completing a task. The literature offers contradicting examples of its benefits; robots trained with social learning reach a higher performance, an increased learning speed, or both, compared to their individual learning counterparts. No general explanation has been advanced for the difference in observations, which make the results highly dependent on the particular system and parameter setting. In this research, we show that even within one system, the observed advantages of social learning can vary between parameter settings. Using Evolutionary Robotics, we train robots individually in a foraging task. We compare the performance of 50 parameter instances of the evolutionary algorithm obtained by a definitive screening design. We apply social learning in groups of two and four robots to the parameter settings that lead to the best and median performance. Our results show that the observed advantages of social learning differ highly between parameter settings but in general, median quality parameter settings experience more benefit from social learning. These results serve as a reminder that tuning of the parameters should not be left as an afterthought because they can drastically impact the conclusions on the advantages of social learning. Additionally, these results suggest that social learning reduces the sensitivity of the learning process to the choice of parameters.

This paper presents an investigation into a population of robots that evolves through embodied evolution an evolutionary process that is not centrally controlled, but emerges from robot interactions just as natural evolution does. The robots select their partners randomly, without reference to any assessment of task performance, but the environment is biased to promote task behaviour by awarding additional life-time to robots that pick up pucks. The experiments show that the robots do learn to pick up pucks in such a setting. Contrary to what one might expect, increasing the amount of additional lifetime awarded decreases task performance for all settings considered. Closer analysis shows that this decrease is in part due to the fact that the increased lifespan decreases the number of opportunities to spread a robots genome, but that increasing the award level also negatively affects selection pressure when there is opportunity for robots to spread their genome. We conclude that higher rewards overly emphasise one aspect of robot behaviour and in doing so prevent evolution from exploring the behaviour space.