May 21, 2025, 13:51 IST /2 min read CEO Jensen Huang steals the show at Computex 2025 with Nvidia's bold AI bets and Taiwan pivot The new clip, some of it played at double speed, suggests Optimus is still catching up to human pace, but the breadth of its skill set is growing. According to Milan Kovac, VP of Optimus (Tesla Bot), Tesla has achieved a “significant breakthrough” in transferring learning from human video demonstrations to the bot’s behaviour, reducing reliance on labour-intensive teleoperation data. The goal now is to expand the learning pipeline to include third-person internet videos and improve performance via reinforcement learning in both physical and synthetic environments. “One of our goals is to have Optimus learn straight from internet videos of humans doing tasks,” Kovac added in his post. The robot’s ability to multitask and adapt to natural language instructions marks a significant milestone in general AI applications. “We’re training a single neural net to do all these tasks, and many new skills are emerging,” said Kovac, adding that Tesla is aggressively hiring for AI roles to accelerate development. Musk, who called Optimus “the biggest product ever” in his reply to the recent video, reiterated in a CNBC interview on Tuesday that Tesla plans to deploy thousands of Optimus units in its factories by this fall. “We expect to scale Optimus faster than any product in history,” Musk said, aiming for one million units annually by as early as 2029. “Everyone’s going to want one.” In a brief exchange with Microsoft CEO Satya Nadella, Musk emphasised that robotic systems, whether robotaxis or humanoid bots, must be grounded in real-world functionality. Trending Stories