A four-legged robot can learn to walk in an hour like a newborn foal

Karmela Padavic- Callaghan is a writer.

A four-legged robot only took an hour to learn how to walk, the same amount of time as a newborn foal.

Felix Ruppert and Alexander Badri-Sprowitz designed a half-metre-high robot that could teach itself how to walk instead of executing a preprogrammed gait.

There isn't much information about the robot's legs, such as the exact shape of each component The neural networks that some animals have in their spine are mimicked by the artificial intelligence.

The instructions are generated by the artificial intelligence. When the robot falls, the readings from the foot sensors are used to adjust them. After about an hour, the artificial intelligence finds the best way to walk.

Because the artificial intelligence learns rather than calculating the details of each leg motion in advance, it can lower the amount of energy it uses as it learns to walk. It walks using less energy at the end of an hour-long learning process.

The way baby animals learn to move is by trying and initially stumbling, which is the most efficient way to use their muscles. You can get energy efficiency for free if you learn to use legs correctly.

Dhireesha Kudithipudi is a professor at the University of Texas at San Antonio. The design of the robot may perform better if it is constantly adjusted.

Hillel Chiel is a professor at Case Western Reserve University in Ohio. Artificial nervous systems are not as complex as they could be.

Adding more sensors and range of motion is one of the things that the team is working on.

Nature Machine Intelligence was published in the journal.