Artificial intelligence helps inform robot’s balance.

A team of researchers, including Australian scientists, have developed a robot that learns to walk across different terrains as it moves.

Autonomous robots have trouble walking, because they typically have a fixed shape and so can’t handle uneven ground or changes in density. This robot can change the shape of its inform legs as needed, and is controlled by an AI that can detect and adapt to different environments.

The robot is described in a paper, published today in Nature Machine Intelligence. In 2018, the researchers released a video of the robot learning to walk.

Just to prove they’re only human, the authors also released a series of “bloopers” – failed robot tests – they filmed during prototyping.

The researchers say their machine is an important step towards commercial walking robots.

“We hope that our research helps pave the way towards flexible hardware platforms that are capable of performing a variety of useful missions in outdoor, unstructured terrains,” write the authors in their paper.