Science Robotics paper learns agile motor skills on ANYmal

Evidence notes

• The paper trains a neural-network locomotion controller in simulation and deploys it on the ANYmal quadruped.
• It reports dynamic motor skills such as fast locomotion, fall recovery, and robustness to pushes, slopes, and uneven terrain.
• This is a landmark ANYmal result because it helped establish sim-to-real reinforcement learning as a practical route for legged-robot control.