Theoretical Foundations for Morphological Computation

The vision of morphological computation proposes that the physical bodies of biological systems, as well of artificial robots, are potential computational resources. The applicability of this vision has already been demonstrated for a variety of complex robot control problems and for a number of different robots. However, a theoretical basis for understanding the capabilities and the limitations of morphological computation has been missing so far. We present two models for morphological computation. They are based on rigorous mathematics and they provide a precise mathematical characterization of the potential computational contribution of a physical body. Remarkably, the presented theories suggest that complexity and nonlinearity, typically unwanted properties of robots, are desired features in order to provide computational power. Based on these models we derive morphological computation devices, which consist of simple generic models of physical bodies, built out of mass-spring systems, and static readouts and/or feedbacks. Remarkably, if the physical body is complex enough, already linear readouts/feedbacks enable such devices to emulate nonlinear, dynamic mappings of input to output streams in continuous time. Hence, by outsourcing the computation to the physical body, the difficult problem of learning to control a complex robot, could be potentially reduced to a simple and perspicuous learning task of finding some linear weights.