July 9, 2024

Investing in Skild

The race to develop general robotic intelligence has begun. In Felicis’ 18 years of operating, no single idea has garnered as significant and as rapid a mobilization of capital and talent. The thing is, the idea isn’t new. Robotics veterans will tell you the holy grail for the industry has always been a single model that can reason, plan, and take action in any environment on any embodiment. The problem with this idea is that it simply wasn’t possible a few years ago. It might be today.

In a world where general robotic intelligence works, it means that scale is (basically) all you need.* It means we can finally address persistent labor shortages for dull, dangerous, and dirty jobs. It’s easy to forget that only 15% of warehouses use any automation, and a mere 5% of warehouses use sophisticated automation equipment. And the majority of industrial robotics installations are concentrated in China and Japan, leaving most of the world far behind. As Jensen Huang, CEO of NVIDIA, recently said, “We’re in the beginning of a new Industrial Revolution. But instead of generating electricity, we’re generating intelligence.”

Enter Skild. When we first flew to Pittsburgh to see the Skild robots in action, we couldn’t believe our eyes. How could a startup in its infancy have achieved so much in such a short amount of time? The answer lies with scale. Large, pre-trained vision language action (VLA) models exhibit the same sort of emergent behavior as large, pre-trained large language models (LLM). Just as training an LLM on Algebra makes it better at Spanish, research suggests training a VLA on navigation makes it better at grasping. Skild is here to push this idea to the max, and we couldn’t be more excited to support them on this journey. 

Our belief in Skild is rooted in the exceptional capabilities of its founders, Abhinav and Deepak. Their collective contributions to some of the world's most advanced robotics and AI labs are a testament to their expertise and vision. We see a convergence of trends that could usher in a golden age of robotics: increasing the volume of quality data, access to compute, and improved underlying hardware (see recent demos from Unitree, Boston Dynamics, Figure, 1X, Optimus, Astribot). We believe a general robotics foundation model will be a crucial link in this chain, breaking the age-old assumption that intelligence is not the bottleneck to successful robotics deployments. The truth is that this assumption is misleading and backward-looking. The field of robotic intelligence has evolved significantly in the last few years. We believe that just as humans learn from visual input—not just text—robots that operate in complex 3D environments hold the key to unlocking physically-grounded intelligence. 

This is a modern space race, and Skild is the team we’re backing.

* This point is a bit more nuanced. It’s worth reading this IEEE Spectrum article.