The field is shifting towards foundation models. There's elegant dynamics to be built, and a universe of robotics to be unlocked. Here’s a rapid tour of what’s coming—and why joining an early‑stage team is a five-way bet on talent-density, future-direction, skill, wealth, and future‑founder credibility.
NVIDIA’s trailer best captures what’s coming:
Google, as they moved from SayCan to RT‑2, reached the same conclusion: more diverse data gives better results than narrow, specialized efforts.
Physical Intelligence doubled down on this insight and have trained Robotic Foundation Models Pi 0 and Pi 0.5:
It’s the bitter lesson—over and over again. While 3D systems might be a necessary stepping stone, especially for validation and reinforcement learning, we’ll probably have to move beyond them.
There's a love of mathematics, physics and hardware at play here. And a lot of good ideas that have become mainstream, espescially since AI made the famous move 37. Let's re-create this magic in robotics!