From the swamp to the screen

A robot that can both walk and swim has been given simulated life in California and researchers believe that understanding the complex behaviour involved in switching from trotting to swimming could lead to a new generation of amphibious robots.

Auke Ijspeert and Michael Arbib of the Brain Simulation Laboratory at the University of Southern California in Los Angeles wanted to investigate how behaviour emerges from simple signals in a creature’s central nervous system. So they built a computer simulation of a salamander’s central nervous system, and superimposed it on computer animation.

The resulting ‘salamander’ exists in a simulated world of flat ground and water. Gravity pulls it down whilst frictional forces act on its feet as it walks, and swirling inertial forces affect it when it is swimming.

Interacting with land and water is a difficult task for a robot, because it has to completely change its gait, and adapt to the new environment, without stopping. Robotic designers have, in the past, tried to solve this by breaking down the problem into parts and solving them individually. But this approach is too inflexible, said Ijspeert.

Animals cope with this problem by using sensory inputs as switches that turn different neural control mechanisms on or off and these are transformed into co-ordinated movements in the body. Through studying salamanders, Ijspeert and Arbib were able to test various ideas about how different neural mechanisms worked and how vertebrates control their bodies.

By copying natural oscillators in the brain that produce rhythmic signals for various types of movement, the researchers were able to produce quite complex behaviours.

‘The circuits are capable of generating trotting and swimming gaits,’ said Ijspeert. When the robot ‘sees’ water approaching, or feels it, a series of neurological switches make it change from the trotting oscillator to the undulating swimming oscillator.