Watch These 10 Robots Walk Like Real People – Thanks to AI Training

The world of robotics just took a giant step forward—literally. United States-based robotics company Figure has made huge progress in teaching its humanoid robots to walk like real people, and the results are seriously impressive. In a new video, Figure showcases its Figure 02 robots walking with a much more fluid and natural motion. Gone are the stiff, jerky movements we are used to seeing. These machines now move with a human-like rhythm, complete with heel strikes, toe-offs, and even synchronized arm swings.

So how did they do it? Figure used reinforcement learning, an artificial intelligence training technique where robots learn by trial and error with feedback through rewards. But here is the game-changer—they did not just put the robot on a treadmill and hope for the best. Instead, they trained their controller in a high-powered, graphics-processing-unit-accelerated physics simulation that ran years’ worth of training data in just a few hours. Thousands of virtual robots, each slightly different, were tested on various terrains and under all sorts of real-world challenges like slipping, tripping, and even being shoved.

What is wild is that all of this learning resulted in one neural network that could control multiple robots without needing to be customized for each one. That means Figure 02 robots can learn fast, adapt easily, and be deployed at scale. In the video, you can see ten Figure 02 robots walking side by side—powered by the same brain.

This breakthrough did not just improve their walking style. It also sped up development, reduced the need for manual tweaking, and opened the door for mass production of intelligent, human-like machines.

So, what does this mean for the future? More lifelike robots, faster development times, and a lot less of that clunky, mechanical robot walk we are used to. Reinforcement learning may just be the key to making robots truly move like us.