Gravis Robotics Patent Portfolio Analysis

Two hundred million, one visible year

Pear VC, IQ Capital and SoftBank put $200M into a Series A for a company founded in 2022. That is a large number attached to a short history. The patent record that an outsider can read consists of three families, all filed in 2024, all still pending. The last public filing is dated 4 October 2024. The round came twenty-two months later.

That gap matters more than the amount. Patent applications publish roughly eighteen months after filing, so anything Gravis Robotics lodged during 2025 or 2026 is invisible for now. The visible portfolio is a photograph of what the company believed in its second year of existence, not what it believes today. Read it as a starting position rather than a finished one.

The file ends where the interesting part probably begins.

Nothing granted, everything still open

Three families published, zero granted. The ribbon on the granted side has no width at all yet. Every claim on file is still a proposal to a patent examiner, filed at the EP and via WO, and first grants are expected around 2027. Nothing in this portfolio is enforceable today, and nothing about it is settled. Scope will be decided in prosecution, not in the applications as published.

The titles show where the arguments will be. One covers methods and systems for controlling autonomous material handling machines. One covers autonomous excavation using a control policy trained with reinforcement learning under depth constraints. One covers a universal control policy for machine actuators. The third is the broadest by name, and the broadest names are usually the ones examiners narrow hardest. Watch what survives.

Four names, one appearing twice over

Four inventors are named across the three families. Fang Nan appears on two. Pascal Egli appears on two. Julian Nubert appears on one. That is a small, tightly overlapping group, which is what you expect when a research team converts a specific line of work into filings rather than when a large engineering organisation files across many fronts.

The concentration reads as coherence. Three applications filed in one year by an overlapping pair of authors, all on control policies for heavy machine actuation, describe one idea pursued in three directions rather than three bets. It also means the portfolio's future depends on a handful of people continuing to produce. Whether the inventor list widens in the 2025 and 2026 filings, once they publish, is the thing to read.

A small circle in a large room

The claims sit in G05B and E02F: control systems, and earth-moving machinery. Over the last five years the same groups drew Fanuc with 592 families, Harbin Inst with 386, Caterpillar with 338, Zhejiang University with 326, Komatsu Mfg with 298 and Bosch Robert with 254. Gravis Robotics has three. On a shared scale, the difference is not subtle.

Count is the wrong measure here, because nobody wins a space like this by volume against incumbents who have filed for decades. What matters is whether the three applications claim something specific that the crowd did not: reinforcement learning under depth constraints is a narrow, technical hook, and narrow hooks are how small portfolios stay defensible in dense art. Crowding raises the bar on scope, not on quantity.

In a crowded class, the useful question is not how many, but how precise.

One citation, and where it came from

So far the portfolio has drawn a single citing organisation: Electronic Sci & Tech China University, with one citing publication. One node, one edge. For applications first published in 2024 that is early and unremarkable, since citation networks take years to form around a young family and only accumulate once examiners and later applicants have the documents in front of them.

What is worth noting is the direction. The one organisation that has cited this work is academic and is working in the same technical neighbourhood. That is consistent with filings that come out of a research lineage rather than out of a product launch. Whether industrial filers begin appearing on that map over the next few years is a real signal about how central these claims turn out to be.

What the hiring plan already tells you

Eighty-three open roles: 35 in software and AI, 30 in hardware, 12 other, 5 in ops and finance, and 1 in go-to-market. Senior Embedded Linux Engineer. Senior Mechanical Engineer. Network Systems Engineer. Technical Program Manager. That is a company still building the machine, not selling it. The shape of the hiring maps closely onto the shape of the filings: control software, actuation, real hardware in the field.

Here is the plain version. The visible moat is three pending families in a crowded class, held together by two inventors, unenforceable until roughly 2027. The $200M was not raised against that file. It was raised against work done since, which will surface on the public record between now and 2028. The 2024 filings tell you the direction. The next publications will tell you the distance.