HANDSWritten for people who want the real answer
What "self-driving" actually means.
The phrase is used to mean about six different things, most of them by people selling something. Here's the version with the marketing taken out.
The levels
There are six, and only one line matters
The important boundary isn't how clever the car is. It's who is legally responsible when it goes wrong — and that changes between Level 2 and Level 3, not anywhere else.
| Level | What the car does | Who's responsible | Real example |
|---|---|---|---|
| 0 | Nothing | You | A 1998 Civic |
| 1 | Steering or speed | You | Adaptive cruise control |
| 2 | Steering and speed, anywhere | You. Always. No exceptions. | Tesla FSD (Supervised) — what we will run |
| 3 | Drives itself in set conditions; you must take over when asked | The carmaker, while it's engaged | Mercedes Drive Pilot, on certain highways |
| 4 | Drives itself with nobody paying attention, inside a mapped area | The operator | Waymo; Tesla Robotaxi in Austin |
| 5 | Drives itself anywhere a person could | The operator | Does not exist. Not close. |
At Level 2 the car can be brilliant, drive for two hours untouched, and you are still 100% responsible for every inch of it. At Level 3 the manufacturer takes on liability while the system is engaged. That is a legal and insurance change, not a technical one, and it is why it moves so much slower than the software does.
How it sees
Eight cameras and no lidar.
Tesla's bet is that if a human can drive on two eyes, a car can drive on cameras alone. Everyone else in the field disagrees. Both positions are still live.
Waymo's cars carry lidar and radar as well as cameras — laser rangefinders that measure distance directly instead of inferring it. It works extremely well and it costs a fortune, which is why Waymo runs hundreds of cars and Tesla runs millions.
The camera-only approach is cheap enough to put on every car sold, which means Tesla collects more real driving data than everyone else combined. That data advantage is the argument. Whether it beats having a sensor that simply cannot be fooled by glare is the open question of this decade.
Our position, since you'd be right to ask: we plan to buy Teslas because they're the only cars we can actually put in a fleet at this price. We are not the right people to settle that argument, and anyone who tells you it's settled is guessing.
- Cameras
- 8 — three forward, two forward-side, two rear-side, one rear
- Radar
- None on these cars
- Lidar
- None. Tesla does not use it
- Ultrasonic
- Removed from this generation; parking is camera-based
- Computer
- Hardware 4 — the minimum that can ever run unsupervised
- Blind spots
- Directly behind the boot lid, and very low immediately in front
Where cameras genuinely struggle
- Low sun straight into a lens
- Heavy rain, spray and standing water
- Snow covering lane markings entirely
- A dirty or salted camera housing
Three of those four are why we expect a worse winter intervention rate than a summer one, and why we have committed to publishing it season by season rather than quoting an annual average that hides it.
August 2026
Where this actually stands right now
Dated deliberately. Anything written about this field goes stale in about four months, and an undated claim is worse than no claim.
Everything in this section is our own reading of public reporting and company statements as of August 2026, not something we verified first-hand and not something either company told us. If any of it is wrong or has moved on, tell us at data@handsfreefsd.app and we'll correct it with a note saying what changed and who caught it.
Tesla runs a genuinely driverless service — in six cities
Austin, Dallas and Houston in Texas; Miami, Orlando and Tampa in Florida. Some of those rides now run with nobody in the car at all. The Bay Area fleet still carries a safety driver because California requires it.
The fleet is small — hundreds of cars, not thousands
Roughly a couple of hundred vehicles across the launch markets as of earlier this year. Impressive as a demonstration, nowhere near a transport network. Waymo has been at this longer and is bigger.
Customer cars are still Level 2, everywhere
The car you or we can buy runs FSD (Supervised). Tesla has stated a target of unsupervised operation in customer cars by the end of 2026, and roughly a dozen states for the robotaxi service. Targets from this company have historically slipped. Plan for it, don't bank on it.
Hardware 3 cars are out
Tesla has confirmed the older computer can't reach unsupervised operation. That's millions of cars that will never be part of this, and it's why our buy rule is absolute.
Some "driverless" rides had a person on the other end
Driverless fleets generally keep remote operators available to help a stuck car, and how much of that happens is rarely broken out in public. That's not a scandal — remote assistance is a normal part of running these services — but it is exactly why we'd rather hand you raw data than a story about it.
Tennessee is wide open, and nobody's here
State law has permitted vehicles without a human driver since 2017 and bars cities from banning them. Waymo is rolling out in Nashville. Nobody has announced anything for the Tri-Cities, and on current patterns nobody will for years. That gap is why this company exists here rather than in Austin.
Six things have to be true before a car of ours can drive with nobody in it. None of them are true yet. This is the checklist, not a progress report.
The plan
Insurance is the bottleneck, not the software.
People assume the blocker is the technology. In practice the software will be ready before anyone will write a policy on it at a price a small fleet can pay.
This is the part nobody talks about. A commercial policy for a vehicle carrying passengers with no driver is not a product most insurers currently sell, and the ones that do are pricing it for companies with a hundred million dollars behind them.
Our answer is the boring one: start building a public record of how our specific cars behave on our specific roads, for years, before we need it. When we walk into that conversation we want to be holding data, not enthusiasm. Right now we are holding neither, which is precisely why the clock starts as early as possible.
That's also the honest reason the telemetry page exists. It's marketing, it's accountability, and it's an insurance file we're building in public. All three at once.
Want to argue with any of this?
Genuinely. If something here is out of date or wrong, tell us and we'll fix it with a note saying what changed and who caught it.