Geckos have the remarkable ability to run at any orientation on just about any smooth or rough, wet or dry, clean or dirty surface. The basis for geckos' adhesive properties is in the millions of micron-scale setae on each toe of the gecko form a self-cleaning dry adhesive. The tip of each seta consists of 100 to 1000 spatulae only 100 nanometers in diameter. Our interdisciplinary team of biologists and engineers has been working since 1998 developing models for how the natural nanostructures function in a hierachical combination of spatulae, spatular stalks, setal stalks, setal arrays, and toe mechanics, and developing nanofabrication processes which allow large arrays of hair patches to be economically fabricated.

The goal of this work is to develop high performance ambulating milli-robots using minimal actuation and passive stabilization mechanisms, combined with onboard high level control.

As the size of a robot decreases, the ratio of its surface area to its volume increases. Because the mass of a robot is proportional to its volume, the increase in this ratio means that surface forces (electrostatic attraction, for example) become large compared to inertial forces. So, as robots (and machines in general) become smaller, friction in their moving parts can become a major source of energy loss, wear, and unpredictable behavior. In the Biomimetic Millisystems lab, we have developed a process called "Smart Composite Microstructures" (SCM) that enables us to build small, strong, lightweight, robots and structures whose ability to move comes from bending of compliant polymer hinges that connect rigid links made from carbon fiber and other composites. These structures are made as single flat pieces and are folded up to form more complicated shapes and linkages. They can also be integrated with smart actuators like piezoelectrics and shape memory alloy to provide motion.

Flapping flight provides the high maneuverability necessary for operation in a partially structured indoor environment. To achieve robust intelligence for tasks such as search and indoor navigation, the maneuverability of the ornithopter will be combined with a learning approach which makes minimal assumptions about the nature of disturbances and obstacles. This approach will develop optimal control policies for single or multiple vehicles. Based
on globally optimal distributed reinforcement learning, we propose to develop algorithms for a set of ornithopters to cooperate in sensing and navigation among unmodelled obstacles such as doors and walls. Our research will be verified with full three dimensional dynamic simulation, a multi-tethered laboratory test-bed, as well as with actual indoor flying ornithopters.