Tesla’s Optimus robot is graduating from dance routines to household chores, as demonstrated in a new video where the humanoid successfully takes out the trash. “I’m not just dancing all day, ok,” quips the robot’s social media account, showing that even artificially intelligent machines can develop a sense of humor about work-life balance.
Behind this seemingly simple task lies a significant breakthrough in robot learning. According to Milan Kovac from Tesla’s AI team, Optimus can now learn directly from internet videos of humans performing tasks. This capability represents a crucial advancement in how robots acquire new skills, allowing them to bootstrap learning from existing human demonstrations rather than relying solely on teleoperated training data.
The Tesla team has developed a system that can transfer learning from human videos to the robots, currently working with first-person perspectives but with plans to expand to third-person random internet videos. This approach mirrors Tesla’s Full Self-Driving technology, which combines real-world recordings with simulations. Skills acquired through this method can be activated through natural language commands (voice or text) and executed by a single neural network capable of multi-tasking. As Optimus continues to take on more household chores, the line between science fiction and reality grows increasingly blurred – though most homeowners would probably prefer their robot assistant begin with a more challenging task than trash duty.