47 experts list 12 things blocking drone fleets by 2035

What happened

You can build a very good car in a country with no roads. It will still be a very good car. However, it will not take you anywhere useful.

That is where drones sit right now, according to a new report. The flying machines are cheap and capable. However, the software, the rules and the shared plumbing behind them are not ready. The report calls this the capability gap.

Therefore, 47 people sat in a room for two days and made a list. Twelve things must be fixed before drone fleets can work at the scale of a road network.

The test

There is no test. This is important, so read it twice. The report is a vision document, so nobody ran an experiment, flew a fleet or measured anything. It records what a group of experts agreed on.

The report is “Computing on the Fly: Navigating a Vision for the Future of Drone Computing”. It is dated July 2026 and was published by the Computing Research Association in Washington DC, through its Computing Community Consortium.

Seven people are listed as authors. They are Kevin Butler of the University of Florida, Christopher Stewart of The Ohio State University, and Nils Aschenbruck of Osnabrück University in Germany. The others are Alina Gerall of the Computing Research Association, Weisong Shi of the University of Delaware, Ufuk Topcu of the University of Texas, and Deborah Silver of Rutgers University. Others contributed too. They include people from Purdue University, the University of Minnesota, the engineering firm Burgess and Niple, and the drone operator DroneUp.

The workshop itself ran on 1 and 2 December 2025, at a hotel in Washington DC. It brought together 47 experts from universities, companies and government. They worked through four rounds of breakout groups over the two days.

The result

The group named 12 technical challenges. Some are about size: scaling to millions of drones, and building reliable fleets from parts that are not predictable. Some are about the machinery in between, such as edge and cloud coordination and the drone networks themselves. Some are about safety, such as security, identity, and guarding key infrastructure. The rest cover data for training and testing, human oversight, standards and rules, and training people to do the work.

The useful part is the 2028 checkpoints, and they are specific. A fleet of 10 to 50 drones finishing 99 percent of missions with no human help, in controlled settings. Safety checking tools that watch a network of up to 100,000 parameters and raise an alert inside 10 milliseconds. Attack detection with a safe fallback inside 100 milliseconds. Twenty or more drones from different organisations working together with no central boss, through 50 percent packet loss and five second blackouts.

The plumbing targets are just as concrete. A national testbed at 10 or more sites. A shared library of over 1,000 hours of failure recordings. Simulations that handle 1,000 or more drones. Training setups at 20 universities.

What it means

Reports like this shape money. This is because research funders and policy staff read them and write programmes from them. If you build in this field, the 12 headings tell you which pitches will find a friendly ear over the next few years.

They also tell you where the holes are. This is because a shared library of 1,000 hours of drone failure footage does not exist. Neither does a 10 site national testbed. Somebody has to build those, and the report says out loud that they are needed.

Read the 2028 checkpoints as product specifications, because they came from people who buy and fund this work. Each one is a gap with a number attached.

Business ideas from this paper

1. Sell a recorded library of drone failure footage, covering sensor faults, bad weather and jamming. Who buys it: drone autonomy teams and university labs who cannot break their own aircraft often enough. A price to test: 1,200 dollars a year per seat. A one-week test: record 10 hours of deliberate failures with one cheap drone, put 30 minutes online free, and email the link to 20 autonomy teams.
2. Sell half-day flight test slots at a private site where customers can deliberately break things. Who buys it: small drone firms who cannot get airspace or permission of their own. A price to test: 2,000 dollars a slot, staff included. A one-week test: get one landowner to agree, list three dates, and see whether anyone books and pays a deposit.
3. Sell a short course on fleet software and the rules that are coming, for people who already fly. Who buys it: inspection and survey firms with pilots but no software staff. A price to test: 400 dollars a seat. A one-week test: run one online session, cap it at 25 seats, and take payment before you build the slides.

How sure can you be?

THIS REPORT CONTAINS NO EVIDENCE, BECAUSE IT WAS NEVER MEANT TO. It is a record of agreement, not a record of measurement. On our scoring it earns nothing for strength of evidence, and that is the correct score for a vision paper.

Every number in the checkpoints is a target somebody set, not a result somebody achieved. The 99 percent mission completion figure is a wish for 2028. No fleet has done it. Therefore, treat those numbers as goals, and never quote them as findings.

The group was assembled, not sampled. Forty-seven people came together for two days. Some of them work for drone companies and engineering firms named in the report. That is normal for this kind of workshop, and it still shapes what gets onto the list.

The view is mostly from the United States. The report says it is written for policymakers, research funders, startups, investors and researchers, and it talks about positioning the United States. Only one listed author is based outside it. What would settle the question is simple. Wait until 2028 and count how many of the checkpoints were met.

Do this today

Open section 4.1 of the report and read the 2028 checkpoints. Pick the one closest to what you already build, and treat it as a written brief from a customer who has not paid you yet.

Source: Computing on the Fly: Navigating a Vision for the Future of Drone Computing, Computing Community Consortium, Computing Research Association, July 2026. arXiv:2607.19213 · arxiv.org (report or dataset).

Just Out Tech explains new research in plain language. This article was drafted with AI assistance and checked by a human against the original source.