Abstract

Morphological computation is a concept based on observations in nature that show that morphology plays a crucial role in the emergence of intelligent behaviour. Our work in the context of robotics demonstrates that morphological properties, such as their dynamic behaviour, can implement computational functionality and, as a result, help to embody control and/or facilitate sensing. This means we can outsource control to the physical body of robots allowing us to build machines with intelligent morphologies. One implication of our work is that, in order to have a computationally powerful body, it needs to have complex nonlinear dynamics, exhibit a high-dimensional state space, and compliance and noise are beneficial. Interestingly, the exact same properties are unwanted and typically suppressed in conventional robotic designs as they make it harder to model and control them. However, counterintuitively, embracing complex dynamics can help us to outsource functionality to the body and therefore simplify the underlying control and learning problems. This points to a paradigm shift in robot design, establishes a potential solution for the control problem of the recently emerged field of soft robotics, and it might provide us with insights into why biological systems are so much more robust and adaptive than our state-of-the-art robots. We will introduce theoretical models for morphological computation and show how they can be used for computation and control in simulations, as well as in real-world robotic platforms. Furthermore, we discuss the wide-reaching implication of morphological computation and how they point to new and exciting research questions leading us from computing octopus arms, to spider web inspired signal processing devices, to artificially growing robots.