“This is the meta-problem of engineering: To design  a machine that can design other machines. Design  automation will be key to the design of complex  systems in domains where there is little formal  knowledge or where human intuition is limited, such  as things on the nanoscale,” says Professor Hod  Lipson in Mechanical and Aerospace Engineering.

Computational synthesis methods attempt to  accelerate and augment the human design process  using machine learning approaches, which learn  through an iterative process of trial, adaptation, and  selection within the virtual world inside a computer  and from data collected from physical experiments  planned automatically.

One practical aim of such research is to  generate machines that are capable of making  inferences and adapting to unforeseen damage or  environmental change.

“Design is a process of searching possibilities  to find an optimal solution, or at least a successful  one,” Lipson says. “If you have 100 components to  work with, then there are literally billions of ways to  arrange them. How do you search that huge space  of possibilities? We are developing the algorithms  for parallel searches to explore the permutations  efficiently.”

When he sees Canada geese flying in formation, Professor Raffaelo D’Andrea in Mechanical and Aerospace Engineering imagines creating feedback control strategies to permit high-flying unmanned vehicles to monitor each other in flight as successfully as birds do so.

“We’re doing fundamental research on the applications of feedback control systems based on mathematical representations of the physical systems they represent,” he says. “When a system has hundreds or even thousands of parts, how do you coordinate them all? The process involves creating subsystems that can interact for the common purpose of increasing performance of the whole,” he says.

One benefit of a fleet of autonomous vehicles flying at, say, 60,000 feet: by flying in an aerodynamically efficient formation, they could stay aloft indefinitely and thus mimic some of the functionality of a spaceborne satellite at a fraction of the cost.