Google Robotics team successfully managed to get a robot to teach itself how to walk without programming its moves. They used a type of machine learning that relies on psychology concepts in order to avoid programming specific commands related to walking, and it worked – the robot basically thought itself how to walk.
Self-taught robotic machines are still undeveloped, which makes this discovery pretty important. It basically proved that the concept works, and you can see how they did it in the video below.
The lead author of the study, Sehoon Ha, published a paper in which he describes how difficult this was. He explained that you can model a road to the tiniest detail, and it still doesn’t help the robot walk down the unfamiliar path. “For this reason, we aim to develop a deep [reinforcement learning] system that can learn to walk autonomously in the real world,” he wrote.
The robot is so advanced, it didn’t even require manual resets during training. The previous robot Ha made in December 2018, for instance, required 100 manual resets.
