Deepak Pathak (left) and Abhinav Gupta (right), former Carnegie Mellon University professors, founded Skild AI in 2023 to build a “general purpose brain” for AI.
Skilled AI
Robots, whether they are bipedal humanoids handling basic factory tasks or four-legged military “robot dogs” for urban combat, require brains. Historically, they have been highly skilled and purpose built. But a Pittsburgh-based robotics startup claims it has created a single off-the-shelf intelligence that can be plugged into different robots to enable basic functions.
Founded in May 2023 by Abhinav Gupta and Deepak Pathak, two former Carnegie Mellon University professors, Skild AI has built a basic model for what it describes as a “general purpose brain” that A variety of robots can be slotted in, allowing them to do so. Things like climbing steep slopes, walking over obstacles and recognizing and picking up objects.
The company announced on Tuesday that it raised a Series A funding round with participation from Lightspeed Ventures, Softbank, Coatue and Amazon founder Jeff Bezos led by Felicity Ventures, Menlo Ventures, Amazon and General Catalyst, among others. Raised $300 million at a $1.5 billion valuation.
Raviraj Jain, a Lightspeed partner who also led the company's seed round in July 2023. Forbes He was impressed with Skilled AI's models when he first saw them being pressure tested last April. The robots they used were able to perform tasks in environments they had never seen before and were not designed for demos. “Back then robots were able to climb stairs, and I think it's really crazy how they could do that because it's a very complex stability problem,” he said.
More impressive, still: Robots using Skuld's AI models also demonstrated “emergent abilities” — entirely new abilities they weren't taught. These are often simple, such as retrieving an object that slips from the hand or rotating an object. But they demonstrate the model's ability to perform unexpected tasks, a phenomenon found in modern synthetic systems such as large language models.
Skilled overcomes this by training its model on a large database of text, images and video – which it claims is 1,000 times larger than the databases used by its competitors. To build this massive database, the cofounders, both former AI researchers, combined a mix of data collection techniques, which they developed and tested over years of research.
One approach was to hire human contractors to operate the robots remotely and collect data about those actions. Another was to have the robot perform random tasks, record the results, and learn from trial and error. The AI model was also trained on millions of public videos.
As a PhD student at UC Berkeley, Pathak developed a way to instill “artificial curiosity” in robots and reward the system for producing results that emerge when they follow their actions. Can't predict the results. “The more uncertain the agent is in predicting the effects of its actions, the more interesting it becomes to explore,” he said, explaining that the technique encouraged the AI to go through more scenarios and collect more data.
He said his research on curiosity-based learning was published in 2017 and has been cited more than 4,000 times. Pathak also devised a way for robots to use written information from large language models such as GPT (how to open a milk carton, for example) and convert it into tasks.
“In 2022 we found a way to put these things into one integrated system,” Pathak said.. “The concept of learning from videos, learning from curiosity, learning from real data but combined with simulated knowledge.”
Skild AI's core model, trained on 1,000 times more data than its competitors, can be slotted into a variety of robots, allowing them to climb steep slopes, walk over obstacles and Enables tasks such as identifying and picking up objects.
Skilled AI
Skild AI faces stiff competition from a string of robotics companies that have emerged with billions of dollars in venture funding thanks to the AI boom. Industry behemoth OpenAI recently revamped its robotics team to provide models to robotics companies. Forbes First reported. Then there are organizations like humanoid robotics company Figure AI, backed by billionaire CEO Brett Adcock, and Covariant, an OpenAI spinoff that's building ChatGPT for robots with $200 million to do so. Has collected more.
Cofounder Gupta claims that Skild AI's access to large amounts of data sets it apart from others in the space, but declined to say how much data its model is trained on.
Ken Goldberg, a professor of robotics and automation at UC Berkeley, agrees that data is the key to measuring robotics, but robots need a specific type of data that isn't widely available on the Internet. . Also, using data collected from simulations doesn't always translate to the real world.
“The whole idea that robotics is excited about right now is that we can do something similar to big language models and big vision language models where they both have Internet-scale data access where you have There are billions of examples.” This is not a straightforward task for robotics, but Skild AI aims to solve this problem by combining all of its data collection techniques with more information from simulations.
Pathak and Gupta envision a future for their company that is similar to OpenAI, where different use cases and products can be built by fine-tuning Skilled's foundational model. “That's exactly how we aim to disrupt the robotics industry,” Gupta said, adding that he would eventually develop artificial general intelligence (a hypothetical AI system that can match or surpass human capabilities) for robots. (can go beyond) want to achieve but one that people can interact with. physical world.
“The world of robotics is having a GPT-3 moment,” said Stephanie Zhan, a Sequoia Capital partner and current investor in Skild AI. “It will create a monumental transformation that we've seen in the world of digital intelligence, in the physical world.”
More from Forbes