Abstract

Animals are remarkable for their ability to move around in complex, cluttered and variable environments. In contrast, most autonomous robots are confined to predictable environments (e.g. roads) or they interact destructively with their surroundings.

This talk will present research on the neural, mechanical and tissue properties of soft animals (such as the caterpillar, Manduca sexta) that explain how adaptive movements are controlled and how this can be related to the design and control of soft robots. The guiding framework is that morphological computation (embodied intelligence) is essential for soft animals to move in the natural world, and that soft machines need to incorporate the same strategies.

Unlike most animals and soft robots, caterpillars do not control their movements through pressure control or pneumatics. Instead they use a tension-based crawling mechanism that allows them to conform to the substrate and remain fully soft. A critical element in this strategy is the ability to control grip release using both mechanical and neural mechanisms to synchronize movements.

Some of these concepts have been implemented in a family of simple elastomeric robots (Softworms) and we have succeeded in generating caterpillar-like gaits based on model-free approaches using motion primitives and through distributed mechanical feedback. The next challenge is to make these robots climb in complex branched structures and to apply the control approaches to more diverse body shapes.