Morphological Modularity Can Enable the Evolution of Robot Behavior to Scale Linearly with the Number of Environmental Features

Abstract

In evolutionary robotics, populations of robots are typically trained in simulation before one or more of them are instantiated as physical robots. However, in order to evolve robust behavior, each robot must be evaluated in multiple environments. If an environment is characterized by f free parameters, each of which can take one of n p features, each robot must be evaluated in all n f p environments to ensure robustness. Here, we show that if the robots are constrained to have modular morphologies and controllers, they only need to be evaluated in n p environments to reach the same level of robustness. This becomes possible because the robots evolve such that each module of the morphology allows the controller to independently recognize a familiar percept in the environment, and each percept corresponds to one of the environmental free parameters. When exposed to a new environment, the robot perceives it as a novel combination of familiar percepts which it can solve without requiring further training. A non-modular morphology and controller however perceives the same environment as a completely novel environment, requiring further training. This acceleration in evolvability -the rate of the evolution of adaptive and robust behavior -suggests that evolutionary robotics may become a scalable approach for automatically creating complex autonomous machines, if the evolution of neural and morphological modularity is taken into account.

Evolving morphology automatically reformulates the problem of designing modular control

“…These results can be used in engineering, in particular, in evolutionary robotics, where they may allow the designers to place a greater number of heterogeneous design variables under the control of evolution, while avoiding, through modularity, the requirement for a combinatorially large number of evaluation environments (Cappelle et al, 2016) and the problem of catastrophic forgetting (Ellefsen et al, 2015) and thus retaining the capacity of evolution to find good-enough solutions in reasonable time. For example, these results may be of interest to engineers and researchers who wish to evolve whole physical agents, complete with morphology, circuitry, and actuator design.…”

“…This coincidence appears analogous to the relationship between modularity and the division approach in engineered systems. In many models, it was indeed observed that evolution produced modular solutions consisting of quasi-independent modules solving subtasks and/or capable of evolving separately (Cappelle, Bernatskiy, Livingston, Livingston, & Bongard, 2016; Clune et al, 2013; Ellefsen, Mouret, & Clune, 2015; Espinosa-Soto & Wagner, 2010; Kashtan & Alon, 2005). This is a non-trivial observation because it has also been discovered that more modular networks, on average, tend to have smaller number of connections than their less modular counterparts (Bernatskiy & Bongard, 2015; Clune et al, 2013; Lipson et al, 2002).…”

Morphological Evolution: Bioinspired Methods for Analyzing Bioinspired Robots

“…For evolutionary roboticists, grand challenges target finding original designs, closing the reproduction loop in physical robots, and allowing for open-ended evolution of physical robots in real environments (Eiben, 2014), which echo the grand challenge from organismal biologists to integrate the analysis of physical and biological systems in order to understand complexity (Schwenk et al, 2009). From this perspective, morphology matters for embodied robots in the same ways that it matters for biological organisms: It permits and constrains individual behavior, and it shapes properties of populations that matter for evolution (Hochner, 2013;Cappelle, et al, 2016). The processes that underlie the evolution of complex morphologies are often themselves complex, for both biological organisms and evolved robots; a deep understanding of evolved morphologies requires techniques to analyze the relevant underlying processes-to answer questions about what morphological forms occur over generational time, and how and why they occur.…”