When most people think of AI and robots, they think of world destruction scenarios involving lasers and glowing red eyes. But Google DeepMind doesn't, and it has taken a major step forward in robotic AI technology by training a system to beat ordinary people at ping-pong.

The video of the robot in action that Google DeepMind posted on X really impressed me as someone with experience programming robotic arms, albeit from a limited manufacturing engineering standpoint.

It's not just the camera tracking the ball and the motors wildly swinging the paddles around. For starters, Google DeepMind has built a database of all the possible initial states a ping-pong ball can have (position, velocity, spin, etc.). From here, the robotic arm practiced various movements and became familiar with how to switch between forehand and backhand grips, apply topspin, etc.

It was then pitted against real players; the AI system was designed to monitor how different people behaved and played, and use that information to refine its overall algorithm. Its success rate was tracked and the chosen strategy self-adjusted in real time accordingly. [...] [...] [...] [...] [...] [...] [...] [...] [...] [...]