Two people familiar with the matter said that Ashish Kumar, former AI lead for Tesla Motors (TSLA) Optimus humanoid robot, has co-founded a startup developing non-humanoid robots to perform tasks in manufacturing, aerospace and other industries that previously required highly skilled technicians.

One of the people said the new San Francisco-based company, named Intelligent Machines, has recruited several former members of the Tesla Optimus team and is currently advancing a seed funding round that is expected to raise about $100 million.

Intelligent Machines is currently focused mainly on robot software development, but management has also discussed developing its own hardware.

According to the two people, the company uses reinforcement learning to train robots: giving positive rewards when the AI achieves a goal and penalties for improper actions, thereby teaching robots to perform complex tasks. (Kumar's doctoral research at UC Berkeley focused on reinforcement learning.)

Reinforcement learning has long been used in robot training, but in the past it was mostly used to teach robots a single skill, such as walking or playing table tennis. Recently, robot developers have begun applying reinforcement learning to more practical tasks, such as rewarding a robot when it successfully inserts a cable or folds a napkin.

One person said Kumar believes existing robot AI models still fall short of the precise control levels needed to automate physical labor in industries such as manufacturing.

It is not yet clear whether Intelligent Machines will adopt vision-language-action models (VLAM). Such models are derived from large language models, the underlying technology behind chat AI, and are specifically trained to control robots. But it is certain that the company's technical approach differs from that of peers focused on world models; world models rely on video data training to understand physical laws and simulate the real-world environments in which physical robots operate. Players in the world model track include General Intuition, which completed a $220 million funding round last month at a valuation of $6.2 billion, and World Labs, which chipmaker AMD plans to acquire for about $8.2 billion.

Although capital continues to pour into robotics startups, AI models for robots have yet to have their "ChatGPT moment" — that is, multiple robots being able to handle a large number of real-world physical tasks in a general way. Venture capitalist Vinod Khosla said in an interview last month that while he believes this moment will arrive in the next two years, he also warned that by 2030 more than half of robotics startups will see their valuations decline.

People familiar with the matter said Intelligent Machines' robot prototype is similar to Optimus in some respects, but it is not a humanoid robot, and there are "major differences" in how it relies on AI compared with Optimus.

Tesla encountered many obstacles in developing the Optimus robot hand, causing mass production delays and constraining the pace of robot expansion. Kumar worked at Tesla for more than two years, including one year as AI lead for the Optimus team, and left one year ago. (His LinkedIn profile shows that after leaving Tesla, he served as a research scientist at Meta for four months.) This summer, Kamal Gupta, a former colleague on the Optimus team, also announced on LinkedIn that he was leaving Tesla.

Intelligent Machines is another startup founded by former Tesla employees. Last month, The Information reported that AI chip company DensityAI, founded by former core members of Tesla's Dojo supercomputing project, was in late-stage talks to raise several hundred million dollars at a post-money valuation of $10 billion.