Tesla thinks it will be able to sell an Optimus robot to every single person on earth, all on its way to selling at least 20 billion of them in total.

Reinforcement learning—the fancy AI industry term for trial and error—was used to train it, including looking at videos of "human motion references" to improve its movements. The result, as you can see, is a robot that can adjust to different shots on the fly and return the shots very naturally for some long rallies.

This week, China’s leading robotics minds gathered in Beijing for a hotly-anticipated, Olympics-style event dubbed the world’s first “ Humanoid Robot Games. ”

This Chinese humanoid robot combines language understanding with a disturbance-resistant gait, pointing to the future of workplace robotics.