.png)
Waymo and Tesla both want a car that can drive itself through a messy American city. After that, they disagree about almost everything. Waymo’s robotaxis pair lidar, radar, cameras and detailed maps, and they already carry paying riders with nobody in the driver’s seat. Tesla’s Full Self-Driving (Supervised) leans on cameras and neural networks, and it still expects a human to be ready to grab the wheel.
As of October 29, 2026, that gap is the whole story. One company sells rides in a growing list of cities. The other sells software to owners who stay responsible for the car. This article walks through how the two stack up on sensors, computing power, busy-street behavior and cost. Everything was checked on October 29, 2026, and we flag the numbers we couldn’t confirm.
The Short Answer
Waymo is ahead on proven driverless operation. Tesla is ahead on potential reach, because its cameras already ride in millions of customer cars. Neither lead is guaranteed to last, and nobody has published independent, head-to-head data that would settle the argument.
- Lidar, radar, cameras and detailed maps
- Paid driverless rides in 14 U.S. cities
- Strong self-reported safety data, plus two big recalls this year
- Cameras and neural networks only
- A human driver stays responsible
- Under a federal engineering analysis in low-visibility crashes
Two Philosophies About Seeing the Road
Waymo’s sixth-generation Driver uses 13 cameras, four lidar units, six radar units and external audio receivers. Each one has a job. Lidar fires laser pulses to build a 3D point cloud. Radar tracks distance and speed, even in rain or snow. Cameras read signs, signals and colors. Waymo’s argument is that the three overlap, so when one is struggling, another can pick up the slack.
That suite is also slimmer than the one it replaces. The fifth-generation setup on Waymo’s Jaguar I-PACE fleet used 29 cameras and five lidars, so the new design cuts the sensor count by 42%. Waymo credits a 17-megapixel imager that sees more of the surroundings with fewer cameras.
Tesla is betting on something different: that cameras plus a powerful neural network can do what human eyes and brains do, with no lidar or radar at all. Its latest notes for FSD (Supervised) v14.3.12 describe an upgraded vision encoder aimed at rare and low-visibility scenarios, along with a retrained reinforcement-learning stage. That’s a software refinement, not new hardware.
Both companies lean hard on machine learning, so the real split is the input. Waymo feeds its models three kinds of sensor data. Tesla feeds one, in enormous volume.
Compute: Tesla Talks, Waymo Mostly Doesn’t
Tesla is fairly open about its hardware. Current cars run the Hardware 4 computer, known as AI4, with 16GB of memory per chip across two chips. An interim AI4+ doubles the memory, and the AI5 chip taped out in April. Elon Musk says one AI5 has five times the compute of two AI4 chips.
| Tesla computer | What’s been reported |
|---|---|
| AI4 (Hardware 4) | 16GB of memory per chip, two chips per car |
| AI4+ | Interim upgrade with double the memory |
| AI5 | Taped out in April; Musk claims 5x the compute of two AI4 chips; volume timing reported anywhere from late 2026 to mid-2027 |
Hardware details come from Tesla-focused coverage and Musk’s own statements, so treat the AI5 claims as promises, not benchmarks.
The timing is murky. Tesla has also said AI5 isn’t initially headed for customer cars, and older Hardware 3 cars get a reduced “Lite” version of FSD.
Waymo hasn’t published comparable specs for its onboard computers, at least not in anything we could find. That makes chip-versus-chip claims guesswork. What Waymo does say is that its savings come from the sensor side: fewer, higher-resolution cameras and cheaper lidar.
Maps: One Side Rejected Them, Then Started Building Its Own
Waymo drives with high-resolution maps alongside live sensing. Tesla argued against that model for years. But on its October 21 earnings call, Tesla’s VP of AI, Ashok Elluswamy, said the company is building its own AI-based maps fed by fleet data, with the first pieces arriving in FSD v15. There was no release date.
Electrek points out this isn’t an HD map in the sense Musk once criticized, since Tesla has always used navigation maps. Still, the gap between “no maps” and “maps” is narrower than the slogans suggest.
Dense City Traffic and Edge Cases
Waymo’s record is strong, but it isn’t spotless. Through its own Safety Impact hub, the company reports 95% fewer serious-injury-or-worse crashes and 82% fewer injury crashes than human drivers across roughly 271 million driverless miles. That data is self-published. A separate July 2026 IIHS study, as reported by the Atlanta Journal-Constitution, found 68% fewer police-reportable crashes per mile.
| Item | What was reported | Who says so |
|---|---|---|
| Serious-injury crashes | 95% fewer than a human benchmark (about 271M driverless miles) | Waymo, self-published |
| Injury crashes | 82% fewer | Waymo, self-published |
| Police-reportable crashes | 68% fewer per mile | IIHS, via the Atlanta Journal-Constitution |
| Waymo recalls | 3,791 vehicles in May (flooded roads); about 3,900 in June (freeway construction zones) | News reports |
| Tesla FSD probe | NHTSA engineering analysis: nine crashes, one fatal, roughly 3.2M vehicles | TechCrunch, citing NHTSA |
We didn’t open the IIHS report itself. The Waymo numbers describe driverless operation only.
Real-world edge cases have still caused trouble. In May, Waymo recalled 3,791 vehicles after an empty robotaxi drove into floodwater in San Antonio. In June it recalled about 3,900 more after more than a dozen incidents of cars entering closed freeway construction zones. Waymo paused freeway service while it fixed the software, and both problems were handled with over-the-air updates.
Tesla’s challenges look different. NHTSA upgraded its probe to an engineering analysis in March, covering roughly 3.2 million vehicles and nine crashes in reduced visibility, such as sun glare, fog or dust. One was fatal. The agency said that in several cases FSD lost track of a lead vehicle or never detected it. An investigation isn’t a finding of defect, and we found no report of a final outcome.
Not a like-for-like comparison: Waymo’s numbers describe driverless operation. FSD (Supervised) has a human fallback by design, so its crashes and saves are shared between the software and the driver.
Tesla’s own robotaxi program shows how early this still is. Unsupervised rides with no in-car monitor began in Austin in January, but Tesla still supervises those cars remotely. Electrek counted only about 42 vehicles in Austin and around 130 in San Francisco in March, almost all with safety monitors. Those figures move fast, and we couldn’t verify current ones.
Redundancy Cost vs. Camera-Only Scale
Extra sensors cost money, though less than they used to. Electrek reported that Waymo’s sixth-generation system could cost under $20,000 per unit on top of the vehicle, more than 50% below the prior generation. Waymo hasn’t confirmed that number. The company also says automotive lidar prices have fallen about 75% since 2021.
Even so, Waymo scales one vehicle at a time. Recall filings suggest a fleet of roughly 3,800 to 3,900 robotaxis, and as of September the company was giving paid driverless rides in 14 U.S. cities. It has talked about 500,000-plus weekly rides and a goal of one million by year-end. That’s an ambitious target, and we couldn’t independently confirm either figure.
Tesla’s counterargument is arithmetic. Cameras are cheap, the hardware is already in customer cars, and every mile can become training data. The catch is that customers still supervise. Elluswamy has said unsupervised FSD for customer cars is targeted for the second quarter of 2027, and Electrek notes that Musk has a long history of missing autonomy deadlines.
Side by Side
| Feature | Waymo Driver (6th gen) | Tesla FSD (Supervised) |
|---|---|---|
| Sensors | 13 cameras, 4 lidar, 6 radar, external audio receivers | Cameras plus neural networks |
| Maps | High-resolution maps plus live sensing | Navigation maps today; AI-built maps planned from FSD v15 (no date) |
| Who’s responsible | The company, in driverless service | The driver, always |
| Where you can use it | Paid rides in 14 U.S. cities (September 2026) | Customer cars, with Hardware 4 getting the newest builds |
| Safety data | Self-published, plus an IIHS study via press report | No Tesla safety statistics verified; NHTSA probe ongoing |
| Cost signal | Reported under $20,000 per unit (unconfirmed) | Hardware already in customer cars |
A simplified comparison as of October 29, 2026. The two companies aren’t selling the same product.
What This Means for U.S. Readers
If you want a driverless ride today, Waymo is the option, in the cities where it operates. If you own a recent Tesla, FSD (Supervised) can take some effort out of driving, but you have to watch the road and be ready to act. Treating the two as direct rivals can hide that difference.
Your Tesla’s hardware matters too. According to Tesla’s release notes, v14.3.12 and v14.3.13 are listed for Hardware 4 vehicles, including the Model 3, Model Y, Model S, Model X and Cybertruck, while Hardware 3 Model 3 and Model Y owners get the pared-down v14.3 Lite. The newer builds also add speed profiles, which pick a speed based on your driver profile, the posted limit and surrounding traffic. None of that changes the legal reality: in a Tesla, you’re still the driver.
What We Still Can’t Verify
- Any apples-to-apples safety comparison between Waymo and Tesla FSD.
- Tesla’s own FSD safety statistics.
- Current Tesla Robotaxi fleet sizes, which change often and come mostly from trackers.
- How the NHTSA probe ends. We found no final decision or recall.
- Waymo’s onboard compute hardware and its exact vehicle count today.
- Waymo’s weekly ride numbers. The 500,000-plus figure and the one-million target are company statements as reported.
The Takeaway
This isn’t simply lidar versus cameras. It’s a question of what a company has to prove before it takes the human out of the car. Waymo paid for redundancy and mapping, and it has driverless miles to show for it, along with recalls that reveal where the hard problems still live. Tesla is wagering that software and scale will close the gap, and its own move toward building maps hints that even a vision-first company sees limits to pure perception.
The next real test arrives with FSD v15 and Tesla’s promised customer-car autonomy. Until independent data exists for both, a healthy dose of caution about every headline number is the smartest position.
Sources
- Waymo: Meet the 6th-generation Waymo Driver
- Waymo: Safety Impact update, September 2026
- Automotive World: Waymo’s 6th-gen Driver and 42% fewer sensors
- BNN Bloomberg: Waymo recalls nearly 3,900 robotaxis
- Drive Tesla Canada: Waymo construction-zone recall
- Drive Tesla Canada: Waymo expands to 14 U.S. cities
- Atlanta Journal-Constitution: Waymo safety study
- TechCrunch: Feds intensify Tesla FSD investigation
- Not a Tesla App: FSD v14.3.12 release notes
- Not a Tesla App: FSD v14.3.13 release notes
- Not a Tesla App: how FSD hardware has changed
- Not a Tesla App: the new AI4+ computer
- Not a Tesla App: Musk on AI5 compute
- Electrek: Tesla building its own maps
- Electrek: Tesla’s unsupervised robotaxi service area
- Ride AI: Waymo launches the 6th-gen Driver
- Images: free to use under the Unsplash License (photographers credited in the captions)
.png){kind=logo}
No comments:
Post a Comment