Taking Shape: A Perspective on the Future of Embodied Cognition and a new Generation of Evolutionary Robotics

Abstract

Abstract The philosophy of Embodied Cognition (EC) presents an intuitive lens with which to consider a variety of problems that attempt to optimise or refine some physically-grounded combination of form and function. Conceptualised as a Venn diagram, with circles of 'body', 'brain', and 'environment', the prevailing notion is that moving to the centre of the diagram is the ultimate goal - intertwining and leveraging all three components to generate adaptive solutions. Philosophically, at least, this makes sense: nature abounds with examples of life that extol the virtues of tightly-coupled embodiment and hint at the possibilities attainable when designing entities through EC principles. The algorithmic basis for following this approach is similarly intuitive: extra degrees of freedom to the design process combined with explicit consideration of the environment allow for a wider range of interesting, useful behaviours. However, moving from philosophy to concrete algorithmic implementation presents a number of pitfalls and barriers that have prevented EC from being more ubiquitously applied as a mainstream problem-solving methodology. In this comment, we speculate on one possible avenue for the future of EC wherein fuller implementations of EC are enabled through adoption of algorithmic advances from the neighbouring field of Machine Learning. Further, we suggest to re-frame evolutionary robotics as a model learning problem, wherein the end goal is to generate an accurate design landscape through the application of high-throughput techniques and tightly coupled digital-experimental systems. Combined, these techniques offer the possibility to reinvent the state of the art and hint at a bright future for evolutionary robotics.