When One Robot Learns, Every Robot Could Benefit

Robots are expanding into streets, farms, factories and industrial sites. Their work also generates large volumes of sensor readings, video, telemetry, diagnostics and operational logs. Moving that data between machines and the computing systems that support them remains a challenge.

The US Federal Aviation Administration expects the commercial drone fleet to grow from about 966,000 aircraft at the end of 2024 to around 1.18 million by 2029. An estimated 4.66 million industrial robots were working worldwide in 2024, and another half a million were installed that year.

Machines need to send their data back to data centers, and may also need maps, software updates, configurations, AI models and instructions in return. The article says that, in 2026, people carrying hard drives remain the fastest way to move large amounts of data off a machine. It argues that network capacity could become a bottleneck as fleets expand and depots handle bursts of uploads and downloads.

Google DeepMind’s Gemini Robotics 1.5 demonstrated “zero-shot” skill transfer between physically different robots: a task taught using a two-armed Franka robot was performed by a humanoid without data being collected on the target machine. The example suggests that connected robots could share what they learn.

The article also points to a 2019 university paper on 6G, which envisioned networks supporting a 10,000-fold increase in traffic. It cites a claim that machines accounted for more than 57% of global web traffic in June of “this year,” and says Cloudflare predicts non-human traffic could exceed human traffic by a factor of 1,000 within five years. The article proposes wireless links as a more flexible way to connect new depots and temporary sites than extending fiber, which can require costly, time-consuming trenching and permits.