Teleo describes itself as a construction robotics startup, but its mission is bigger than automating heavy equipment like excavators and tractors. Today, Teleo’s retrofitted machinery allows its customers to operate their existing fleets semi-autonomously. In the future, the startup sees the data it collects as a key enabler for the robotics industry to reach its “ChatGPT moment.”
That isn’t an aspiration to reach the same level of hype surrounding ChatGPT. Instead, Teleo CEO Vinay Shet sees an opportunity for robotics companies — and namely the one he runs — to gather vast datasets similar to amounts used to build ChatGPT in order to make big, game-changing leaps in robotics.
And investors seem keen to help the startup reach that milestone. TechCrunch has learned that Teleo recently raised $16.2 million in funding through two extensions to its 2022 Series A round. The $9.2 million extension closed in April and another $7 million one closed this week, according to recent filings and information from the company.
“The foundation models that led to the creation of ChatGPT relied significantly on, effectively, trillions of tokens worth of data that was freely available on the internet, on language, on videos, on images, and so forth. That data does not exist in robotics,” Teleo CEO Vinay Shet told TechCrunch. “The best data set that we know of in the robotics world is about 2.4 million tokens, whereas in the language world, they train it on trillions of tokens.”
Teleo aims to bridge that gap by logging data from its own daily operations, which Shet says will end up “becoming the basis upon which you can train true robotic foundation models” that can lead to generalized intelligence.
To build that repertoire of data, Teleo needs to deploy rapidly, at scale, and across several industries. And the company’s strategy for doing that is wrapped up in its semi-autonomous approach. Teleo can retrofit any piece of equipment with the necessary self-driving software and sensors – like cameras, lidar, and radar – to drive themselves autonomously in limited conditions. Remote human operators then step in to perform more complicated tasks, like unloading a dump truck, and can typically handle multiple vehicles at a time.
“That combination is what lets us solve the whole use case for the customer while delivering [return on investment] to the customer and making money through a standalone product,” Shet said.
In order to maintain a diverse dataset, Teleo recently expanded beyond construction and is deploying autonomous heavy machinery, like wheel loaders, terminal tractors, and excavators, across a range of industries, including pulp and paper, logging, port logistics, agriculture, and munitions removal. Teleo is also targeting industries like airports, waste and recycling, logistics, and snow removal.
The hope is that the data it collects – including inputs from human operators, video footage, and sensor feedback – will allow Teleo to fine-tune or specialize foundational robotics models. This could eventually enable the replacement or augmentation of the human in the loop with a cloud-based AI agent capable of learning to control different machines, as a human would.
This long-term thinking is no doubt what attracted investors to the company. Teleo’s recent extensions were led by UP.Partners with participation from new investors Trousdale Ventures and Triatomic Capital, as well as returning investors F-Prime Capital and Trucks VC, among others.
Teleo says the funds will be used to scale customer deployments, continue expanding to new industries, and enhance the startup’s AI capabilities, including an integration of large language models (LLMs) to unlock operator efficiency.
“Over the next several years, you will see vertically integrated companies like ourselves actually deploy in the real world in a manner that makes sense economically speaking and grows economically based off that,” Shet said. “But along the way, they’ll collect enough data in the right format so that it unlocks that ‘aha’ moment a few years down the road.”