Survival of the fittest

Robots present an enduring image in popular culture but building one requires teams of specialists, a lot of money and a whole lot of time.

However, researchers at Brandeis University have developed a method where a robot can design other machines with the minimum of human input.

Hod Lipson, a mechanical engineer, and Jordan B Pollack, a computer scientist, claim their work represents a new step towards the ‘autonomy of artificial life’.

Electro-mechanical systems driven by neural works were evolved inside computer simulations to yield robotic blueprints that can be manufactured automatically by a rapid prototyping machine.

Lipson envisioned and designed the software to simulate and evolve truss structures. Fixed points were replaced with ball joints, and variable length motors driven by artificial neurons acted as the brain and muscle tissue. This gave Lipson and Pollack an accurate simulator for the machines operating much faster than real time.

Evolutionary computation provided methods for searching through the myriad of machines using variation and selection. Simulated machines were varied by changing the connectivity and weight of neurons, the location and length of bars, and were selected for reproduction based on their locomotive ability.

The computer software took a few days to construct bodies-with-brains that best optimised the mobility criteria for survival. Each run resulted in the creation of a diverse locomotive mechanical and control circuit without human input.

A rapid prototyping machine gave ‘life’ to the machines by changing points and lines inside the computer into ball-joints and cylinders, giving Lipson and Pollack a geometric blueprint of a machine that could be constructed and recycled for approximately $250

Lipson and Pollack see future experimentation involving more complex environments and sensors so robots can react to obstacles and other robots.