Constrained by Design: Influence of Genetic Encodings on Evolved Traits of Robots

Abstract

Genetic encodings and their particular properties are known to have a strong influence on the success of evolutionary systems. However, the literature has widely focused on studying the effects that encodings have on performance, i.e., fitness-oriented studies. Notably, this anchoring of the literature to performance is limiting, considering that performance provides bounded information about the behavior of a robot system. In this paper, we investigate how genetic encodings constrain the space of robot phenotypes and robot behavior. In summary, we demonstrate how two generative encodings of different nature lead to very different robots and discuss these differences. Our principal contributions are creating awareness about robot encoding biases, demonstrating how such biases affect evolved morphological, control, and behavioral traits, and finally scrutinizing the trade-offs among different biases.

The Effects of the Environment and Linear Actuators on Robot Morphologies

“…This is further visible in Figure 4, where we visualize the morphologies of the best individuals from both populations. The majority of the resulting morphologies for this experiment are 'snake'-like as observed in previous works [26,27,39]. In these 'snake'-like morphologies, swapping some active hinges for linear actuators does not change its behavior as is further visible in the video, explaining why the linear actuator still gets selected but does not affect the fitness.…”

“…These results are surprising since they contradict the main principles of evolution. Further investigation revealed that the 'snake'-like morphologies were the result of a bias in the reproduction operators of the used genetic encoding, namely the L-System [26].…”

Morphological Evolution: Bioinspired Methods for Analyzing Bioinspired Robots

“…Along with that capacity comes a need for caution, however, to avoid the possible experimenter bias that can accompany a detailed knowledge of the internal representations. Because biologists do not have that window into natural, living organisms, the selection gradient and morphospace walk methods in this paper rely only on externally observable traits and properties; this is consistent with some other morphological analysis of evolved robots (e.g., species determination in Medvet et al, 2021), but it reflects a complementary perspective to analytical approaches founded in knowledge of underlying genetic encodings (e.g., Miras, 2021).…”

“…Indeed, the command of variables and values afforded by the computational-robotic paradigm enables more detailed explorations than could be achieved with living biological organisms, but because roboticists design the algorithms for their systems, they assume the extra responsibility for understanding potentially unintended effects of their implemented processes-including, but not limited to, unintended biases in development algorithms. One of our goals in presenting the analytical methods in this paper is to enable bioroboticists (and other scientists and roboticists) to analyze the effects of their designs from a basis of observed behavior and morphology, an alternative to approaches that intrinsically focus on encodings (e.g., Rothlauf, 2002;Miras,…”

“…For example, we do not in this paper explore differences among encoding schemes (cf. Miras et al, 2018;Veenstra and Glette, 2020;Miras, 2021), environments (cf. Auerbach and Bongard, 2014), morphological descriptors (cf.…”