Robotic System

3/564 → 15/777; +1.40 percentage points of formation share.

What is striking about this batch is how few of these companies are actually building robots. Lightberry, Avea Robotics, Instance, and Waddle Labs are building the software layer that sits on top of hardware someone else already made, from voice control to uptime monitoring to automated evaluation, while only a handful, like Origami Robotics with its own hand-based data-collection device or Philon's open-source general-purpose robots, are touching physical design at all. The rest, from Manifold to InLoop Robotics to Remy AI, are selling labor outcomes to warehouses and 3PLs rather than machines, which tracks with Manifold's own framing that the $75bn warehouse labor market has stayed just 10% automated because the hardware has been too expensive to justify. That gap between fast-moving control software and slow-moving physical deployment explains why this formation's 3.63× growth trails Industrial at 8.98× and Hardware at 5.49×: the surrounding infrastructure and factory demand are scaling faster than the robotic systems meant to serve them. The timing lines up with the arrival of capable, cheap vision-language-action models over the past two years, the same wave visible in the new Hugging Face clusters around Pi05, Groot, and OpenPI, which finally made it plausible for small teams to teach a robot a task from a single demonstration, as Shiraz AI is doing, instead of hand-coding it. That shift also explains the emergence of companies like Instance, whose entire business is evaluating robot policies, a job that only exists once policy-learning robots are common enough to need grading at scale.

No comparable outside activity is connected yet.