White paper · Public edition
Cede: taking turns between robotaxi companies
A short read on what Cede is for, how it works in plain terms, what it promises, and how far it has been tested. This is the limited edition. It leaves out the specification, the parameters, and anything you would need to build Cede, because those are proprietary.
The short version
- Two robotaxis from rival companies meet at a junction with no traffic lights. Each waits politely for the other. Nobody moves.
- Cede is a shared rule that says who goes first. The car still drives itself; Cede never touches the steering or the brakes.
- A waiting car says only where it is and that it's stuck. Not where it's going, and not who's inside.
- Both cars reach the same answer for whose turn it is, and neither company can tilt it.
- So far it lives in simulation: more than a billion tests, and no public roads yet.
- Cede is proprietary. This paper is here to explain it, not to license it.
1. The problem
Two robotaxis come to a junction with no traffic lights, on paths that cross, and both stop. Each one sees the other sitting there, apparently yielding. Each decides it can go only if the other keeps waiting, so it waits too. Both cars have read the scene correctly. The standoff is in the decision, not the sensors, which is why a sharper camera or a better radar does nothing for it.
Inside one company this is a non-problem: the operator just tells its own cars what to do. Between rivals it is a real one. They run different software, answer to no shared referee, and none of them will take a competitor's word for how to drive. Today the only way out is for one car to creep forward and hope the other yields. That is slow, it looks nervous from the kerb, and it gives no company a fair shake.
2. How big the problem is
Driverless taxis stopped being a pilot project some time ago. In May 2024 Waymo was giving about 50,000 paid rides a week. By March 2026 it reported 500,000 a week, across ten US cities.
Waymo no longer has those streets to itself, either. In at least nine US cities more than one robotaxi fleet now shares the road, counting services open to the public and ones still limited to staff or run with someone at the wheel. San Francisco has four.
The cars do get stuck, and it is on the public record.
None of these is the exact case Cede is built for. Most involved one company's cars, or cars that stopped for some other reason. Nobody has counted how often two rival cars freeze at the same junction, each waiting for the other. What the record does show is that stalled driverless cars are a real problem regulators already act on, and that the list of cities where rivals meet at the same junctions keeps getting longer.
3. How Cede resolves it
Cede is about the smallest thing two cars could say to each other and still break the tie. A waiting car broadcasts a short, signed message that amounts to “I'm here, and I'm stuck.” It carries where the car is and the fact that it's waiting. It does not carry where the car is going.
From there, both cars work out the same answer to one question: whose turn is it? That answer rides on identities that neither car picked and neither can guess ahead of time, so no company can set itself up to win. Nothing in Cede says “you may go.” The car whose turn it is still checks the way is clear with its own sensors, then goes. The other sees it move and follows.
Since the cars trade no intent, no company has to believe what a rival claims, and there is nothing worth lying about on the air in the first place.
Published standards already let cars coordinate, and they do it by having each car tell the others what it plans to do. That suits cars whose makers trust each other. Cede goes the other way and sends no plans at all, which is the reason rivals can use it.
4. Who stays in control
Cede never steers and never brakes. All it hands the car is an answer to “whose turn is it?”, and the car's own driving system can take that answer or leave it. The car keeps watching the road the way it always did. If something looks wrong, or if the Cede messages go missing or arrive garbled, the cars act exactly as they would with no Cede at all. The one thing Cede can do is give a car a reason to wait. It can never give one a reason to go.
6. What it promises
- Nothing private is shared. A car says where it is and that it's stuck. Anything more goes only to partners it chose, locked. Never who's inside.
- No one is in charge. No central server, no referee, and no company that owns the decision.
- Every car keeps its own judgment. Cede never steers or brakes; each car still checks the way is clear itself.
- Every turn is fair. A shared result that neither car nor company can influence.
- When in doubt, it steps aside. If anything is unusual, the cars behave as they would without Cede.
- Anyone can check the result. Each car keeps a signed record, so a third party can confirm who should have gone first.
7. The evidence so far
Every number below comes from simulation. None of it is measured on real vehicles, and none of it shows that Cede is safe to run on real roads.
- More than a billion tests. 1,145,499,391 simulated meetings, some of them with cars built to misbehave. The runs that ended unsafely turned up nine flaws. All nine are fixed, and the final version has not produced one since.
- Fair over the long run. Across 610 million simulated tosses, each company went first about half the time. The gap between any two companies came out to a fraction of a percent.
- Tried inside self-driving software. Cede has run inside simulated self-driving systems with all of their own safety checks left switched on.
- Quicker to clear. In simulated city traffic, standoffs broke in about 6 seconds with Cede and about 11 without it, and without the stop-start shuffle of two cars edging out and backing off.
8. What is not yet proven
Cede is still a proposal. No authority has certified or approved it, and it is not on public roads. What comes next is bench work and field work: measure how far the “I'm here” message actually carries between real radios, then two cars on a closed track with safety drivers, then a quiet watching period on real fleets to see how often these standoffs even happen, and only after all that a first supervised service. Nobody has measured yet how common or how costly these cross-company standoffs are on real streets. That measurement, not this paper, is what the case for deploying Cede will stand on.
9. Ownership
Cede is proprietary. Its specification, its software and its test results are owned, and all rights are reserved. This paper explains the idea for the public; it is not a licence. Nothing here lets anyone use, copy or build Cede, and it leaves out the specification, the parameters and the working detail on purpose. For the full specification or to ask about licensing, write to the address below.
10. Contact
Questions about Cede, the full specification, or licensing go to contact@autonomousstandard.org.
Using this website is covered by our terms of use, and what we do with personal data is set out in our privacy policy.
11. Sources
Outside figures were read on 30 September 2026. Cede's own figures come from its simulation campaign.
- Waymo ride counts: TechCrunch, March 2026; CNBC, December 2025; CNBC, April 2025; The Verge, October and August 2024.
- Fleets by city: The Charge Port robotaxi tracker; Smart Cities Dive.
- San Francisco fire department reports: Mission Local, August 2023.
- Cruise cars stalled in North Beach: The Independent, August 2023.
- Cruise permits suspended: California DMV, 24 October 2023.
- Waymo investigation: NHTSA, PE24-016; Reuters, July 2025.
- San Francisco power cut: BBC; Fortune, December 2025.
- Standards: SAE J3186; ETSI TS 103 561.