What do companies study in robotics?

In 2019 engineers at China Southern Power Grid, one of China’s two grid operators, published a paper on a chore that sounds nothing like robotics: inspecting high-voltage transmission towers. The paper works out safety rules and flight routes for a small drone to fly the tower and photograph it automatically, tower after tower. That is a robotics paper by any measure we use: it sits among 3,853 papers that cite one another on the same problem, which we file as one research topic. It has been cited 81 times. And its authors work for a power company, not a university or a robot maker. This story asks how often that happens, and where.

Three problems, three kinds of company. Driving — 78,776 papers on how a car sees the road and decides — is where the car industry publishes: Daimler (Germany), Robert Bosch (Germany), Ford Motor Company (United States). Learning — 55,019 papers on how a machine improves by trial — is the technology companies’: Google (United States) first by a distance, then Meta (Israel). Inspecting power lines — 3,853 papers — is the utilities’: China Southern Power Grid (China) and State Grid Corporation of China (China). Two denominators, and they answer different questions. The bars count papers in a topic that name a company; they do not say how a company spreads its papers across topics.

Over the last ten years, 11.7% of the driving papers we can attribute carry a company author, 11.9% of the learning papers, and 20.9% of the power-line papers — against 6.1% across every topic we place. The average is small because most robotics research is academic; inside it, three very different stories run. What the lines do not say: a company author is a company taking part in research, not a company using the technique, and not a product. And a line can move for a dull reason — the share of papers whose authors name any institution differs by topic and by year — which is why the caption states it.