Writing‎ > ‎Science news‎ > ‎

Controlling Chaos

A new six-legged robot invented by German scientists could help make the dream of autonomous machines true.

Researchers from the Bernstein Centre for Computational Neuroscience have developed a robot that decides how to move using a single, central neural network. To do this, it gathers information from eighteen sensors and attempts to convert these inputs into sensible motion. The hexapod features in this week’s issue of Nature Physics.

Every time you feel the weather is icy and your foot slips, you automatically walk more tentatively. As conditions improve, your walking changes back to normal. This is possible thanks to an internal central pattern generator (CPG), which takes sensory data, analyses it, and generates an appropriate response: in this case, the correct movements.

Other walking robots have a number of CPGs and choose the appropriate one depending on sensor inputs.

“Whereas typically robots have one CPG for each gait... our chaos controlled robot needs solely a single control unit,” explains lead researcher, Silke Steingrube. This chaos control lets the robot behave more like a living creature, adapting quickly to its surroundings.

Chaos might sound random, but it actually refers to systems that produce radically different outputs from subtlety different inputs.

“We developed a tool to control this chaotic CPG to different target periods, such that the robot would turn from chaotic to regular motion,” adds Steingrube. Without stimuli from the sensors, the robot moves wildly. When the scientists include sensor information, however, it narrows its actions to a single periodic walk.
Dr. Stephen Roberts, head of Machine Learning at Oxford University, said the work was “interesting, but not advanced enough to see how it could outperform conventional control theory”.

The real advantages are in the robot’s versatility. Because the robot acts on the brink of chaos, changes in sensor input cause it to change gait almost instantly, and if it gets stuck in a hole, the robot uses its chaotic movements to escape.

Building on their chaos control, the researchers’ next move is to provide the robot with motor memory, so it can plan ahead and act with greater autonomy.