Skild AI grabs $300M to build foundation model for robotics

A new wave of startups is emerging to apply recent artificial intelligence advances into more sophisticated robot behavior. Yesterday, Skild AI emerged from stealth mode and announced that it has closed a $300 million Series A round. The Pittsburgh-based company said the funding brings its valuation of $1.5 billion.

Skild AI is chasing one of the most significant goals in current technology: embodied intelligence — creating an intelligent device that can learn and react to the world on its own. The company is developing the Skild Brain, a robotics foundation model, as well as a mobile manipulation platform and a quadruped platform for security and inspection.

There are primarily two approaches to developing an embodied AI solution. Each approach has its pros and cons.

Vertical approach

With a vertical market approach, you find a market problem, design a mechanism to solve the problem, and train a bespoke AI brain (or model) to operate the mechanism. This might also be called the specialist approach, and it is the approach that many robotics startups have taken.

This approach lowers the risk and provides more control over the outcomes, plus complete control of the intellectual property. In the current humanoid robot race, each participant is trying to design a unique mechanism and embodied intelligence.

Horizontal approach

With a horizontal market approach, you create a broadly intelligent system that is capable of learning any task and then make it capable of being deployed to control any mechanism. This is the path to artificial general intelligence (AGI), but it could also be called the foundation model approach, as a foundation model can serve a general class of problems while not being a vast AGI. This is the path that Skild is taking.

Skild AI joins companies such as OpenAI, Covariant, and NVIDIA that are attempting to build a foundation model that can be applied to multiple applications, irrespective of the mechanical design of the mechanism.

Register now.

Why Skild AI might be different

Skild AI claimed it is training its model on at least 1,000 times more data points than competing models. As opposed to vertically designed robots that are built for specific applications, Skild’s model would serve as a shared, general-purpose brain for a diverse embodiment of robots, scenarios, and tasks, including manipulation, locomotion, and navigation.

From resilient quadrupeds mastering adverse physical conditions to vision-based humanoids performing dexterous manipulation of objects for complex household and industrial tasks, the company said its model will enable the use of low-cost robots across a broad range of industries.

“The large-scale model we are building demonstrates unparalleled generalization and emergent capabilities across robots and tasks, providing significant potential for automation within real-world environments,” stated Deepak Pathak, co-founder and CEO of Skild AI. “We believe Skild AI represents a step change in how robotics will be scaled, and has the potential to change the entire physical economy.”

In contrast to existing robotic applications, where robots are deployed in isolated or constrained environments, Skild’s general-purpose AI model is intended to make any kind of robot agile, dexterous, and safe to interact with people. The market opportunity for embodied AI is vast, and as a result, investors are now betting large sums on the race to develop both foundation models and vertical robotic solutions.

Co-founders have AI credibility 

Skild is in good company, with organizations such as Sanctuary AI, Figure AI, Robust AI, and Brightpick AI obtaining large investment rounds over the past two years.

Co-founders Pathak and Abhinav Gupta left Carnegie Mellon University (CMU) professorships to start Skild. The pair has a combined 25 years of experience in robotics and AI, and they have been credited for numerous industry breakthroughs such as self-supervised robotics, curiosity-driven agents, and adaptive robot learning.

Together, they have a 150+ h-index, over 90,000 citations, and have received multiple awards for their work. The Skild AI team includes robotics and AI experts from Meta, Tesla, Nvidia, Amazon, Google, and schools including CMU, Stanford, and UC Berkeley.

“With general purpose robots that can safely perform any automated task, in any environment, and with any type of embodiment, we can expand the capabilities of robots, democratize their cost, and support the severely understaffed labor market,” said Gupta, co-founder and president of Skild AI.

Tech heavy hitters bet on Skild AI

Lightspeed Venture Partners, Coatue, SoftBank Group, and Bezos Expeditions led Skild AI’s Series A round. Other participants included Felicis Ventures, Sequoia, Menlo Ventures, General Catalyst, CRV, Amazon, SV Angel, and CMU.

“Skild AI has achieved massive breakthroughs in a short period, and we believe they’re a one-of-a-kind company that could redefine our notions of what machines are capable of,” said Raviraj Jain, a partner at Lightspeed Venture Partners. “Deepak and Abhinav have been catalysts of advancements in robotics, and their innovation around leveraging the core principles of foundation models into the real world puts the industry on the path of general-purpose robotics.”

“A GPT-3 moment is coming to the world of robotics,” said Stephanie Zhan, a partner at Sequoia Capital. “It will spark a monumental shift that brings advancements similar to what we’ve seen in the world of digital intelligence, to the physical world.”