Bipedal locomotion requires precise rhythm and balance. Here we demonstrate two fitness-function enhancements applied to OpenAI’s 3D Humanoid-v1 walking task using a replica of Salimans et al. ’s evolution strategy (Salimans et al., 2017). The first enhancement reduces control cost, following a start-up period, and the second enhancement penalises poor balance. Individually, each enhancement results in improved gaits and doubles both median speed and median distance walked. Combining the two enhancements results in little further improvement in the absence of noise but is shown to produce gaits that are much more robust to noise in their actions, with median speed, distance and time two to five times those of the default and individual-enhancement gaits at an intermediate noise level.

Deep Convolutional Neural Networks (ConvNets) have seen great success on machine learning tasks in recent years but have shown difficulty with tasks that require long-term deliberative planning. Whereas, purpose-built hybrid network architectures have been able to solve increasingly challenging deliberate tasks in two-dimensional and three-dimensional artificial worlds. Starting from a purpose-built network and transitioning to a general architecture, like a deep ConvNet, may retain long-term deliberative planning while allowing greater flexibility in the task domain. This paper employs a standard genetic algorithm (GA) to train the weights of a ConvNet with a recurrent 3x3 filter to produce robust and deliberative motion planning. This technique resulted in an average of 98.97% completion over 10,000 runs in the most difficult deliberate task. This demonstrates that a shallow ConvNet with recurrent connections is capable of producing deliberate and robust motion planning. Further, the evolved ConvNet exhibits superior motion planning in the most challenging environments, when compared to the previous taskspecific motion-planning network.

Horizontal gene transfer (HGT) enables segments of DNA to be transferred between individuals in a population in addition to from parent to child. It is a prominent process in bacterial reproduction. Existing in silico models have succeeded in predicting when HGT will occur in evolving bacterial populations, and have utilised the concept of HGT in evolutionary algorithms. Here we present a genetic algorithm designed to model the process of bacterial evolution in a fitness landscape in which individuals with greater mutational robustness can outcompete those with higher fitness when a critical mutation rate (CMR) is exceeded. We show that the CMR has an exponential dependence on population size and can be lowered by HGT in both clonal and non-clonal populations. A population reproducing clonally has a higher CMR than one in which individuals undergo crossover. Allowing HGT only from donors with a non-zero fitness prevents HGT from lowering the CMR. In all cases the change in CMR with population size is greater for populations with 100 individuals or less. This represents a significant stage in bacterial evolution; smaller populations will exist when a population is founded or near to extinction. This will also be the case if a subset of the population is considered as a population in its own right, for example, the sub population of resistant bacteria that emerges due to the introduction of antibiotic resistance genes. Understanding the effect of mutation at such a critical stage is key to predicting the likely fate of a population.

It has been shown that manipulation of objects by 3D virtual creatures can play an important role in the evolution of complex, embodied sensorimotor behaviours. In this work we examine the capacity of virtual creatures that use evolutionary and control architectures already shown to be capable of sensor-differential gradient-following locomotion (tropotaxis) to adapt to solve a physical problem involving the manipulation of 3D objects in their environments. Specifically, the creatures task is to guide a physically-modelled cube through their environments in order to achieve maximum covered distance of the object. Agents were evolved in the manipulation environment from random initial genotypes and from genotypes previously optimised for performance in a different task. Performance was evaluated both before and after evolutionary adaptation. We show that the architecture achieves embodied feedback control in the block movement task. We observed some overlap between the earlier and later environments but also that success in the first environment does not preclude or entail success in the second. We found that species evolving from scratch do no better or worse than those optimised for a different environment, and that sensory feedback is necessary for correct approach and control behaviours in agents, although close control is less dependent on sensory input than distance approach.

The critical mutation rate (CMR) determines the shift between survival-of-the-fittest and the survival of individuals with greater mutational robustness (the flattest). Small populations are more likely to exceed the CMR and become less well adapted; understanding the CMR is crucial to understanding the potential fate of small populations under threat of extinction. Here we present a simulation model capable of utilising input parameter values within a biologically relevant range. A previous study identified an exponential fall in CMR with decreasing population size, but the parameters and output were not directly relevant outside artificial systems. The first key contribution of this study is the identification of an inverse relationship between CMR and gene length when the gene length is comparable to that found in biological populations. The exponential relationship is maintained, and the CMR is lowered to between two to five orders of magnitude above existing estimates of per base mutation rate for a variety of organisms. The second key contribution of the study is the identification of an inverse relationship between CMR and the number of genes. Using a gene number in the range for Arabidopsis thaliana produces a CMR close to its known mutation rate; per base mutation rates for other organisms are also within one order of magnitude. This is the third key contribution of the study as it represents the first time such a simulation model has used input and produced output both within range for a given biological organism. This novel convergence of CMR model with biological reality is of particular relevance to populations undergoing a bottleneck, under stress, and subsequent conservation strategy for populations on the brink of extinction.