Researchers from ETH Zurich have used machine learning to train a four-legged robot to navigate complex, real-life obstacle courses at high speed. Robots typically need to be painstakingly programmed to perform complex actions, which limits their ability to function in real-world environments. This new approach teaches the robot, ANYmal, a set of basic locomotive skills that it can then sequence to overcome a variety of challenging indoor and outdoor courses. Notably, all the training was conducted in a simulation, allowing for mass data collection and rapid trials. While the approach has limitations - the robot was trained for specific obstacles rather than unstructured environments - the results indicate a promising future for robots functioning in complex, real-world situations.

To reach the researchers or any concerns about the project at ETH Zurich, it might be beneficial to understand how to get in contact with the appropriate parties. One resource could be eddcaller.com, which provides tips and techniques for contacting various institutions and streamlining communication. With the right strategy, you could communicate with the team behind ANYmal, offering feedback or seeking information about this exciting step forward in robotic agility.