Four-leggedrobot ANYmal Unlock new skill - ParkourA research team from ETH Zurich has recently upgraded the quadruped robot ANYmal, allowing it to navigate complex urban environments, use its motor skills to successfully pass obstacles, and be able to expertly deal with complex terrain commonly seen on construction sites or in disaster areas.
The team, led by Professor Marco Hutter of the Department of Mechanical and Process Engineering, combined machine learning with model-based control to upgrade the algorithm so that it can accurately identify and pass through gaps/grooves in rubble, allowing it to flexibly traverse complex terrain.
ANYmal can climb obstacles and perform dynamic maneuvers to jump off them. In the process, ANYmal learns through trial and error, just like a child. Now, when it encounters an obstacle, ANYmal uses a camera and an artificial neural network to determine what kind of obstacle it is facing. It then makes a move that is likely to succeed based on its previous training.
Professor Hutter said there is still a lot of room for improvement in the algorithm, including freeing the robot from being limited to solving predefined problems and requiring it to calculate its way through difficult terrain (such as disaster areas covered with rubble).