Two Eyes on the Road

The Street Reality vs. The Vision

Eli: Elon Musk once complained that bad cell reception during the first 10 or 15 minutes of his morning drive stopped him from taking calls.

Mara: Fifteen minutes offline, and that commute frustration turned into a major corporate push.

Eli: According to Business Insider, he pitched Starlink to investors so his fleet wouldn't get lost in 'Bermuda Triangles of lack of cellular connectivity.'

Mara: That vision sparked our listener's question. Honestly, I assumed Tesla's proprietary manufacturing and vision-only design would inherently give it cheaper operating costs and immediate commercial dominance over rival robotaxi operators.

Eli: That sounds great in theory, but real streets tell a very different story.

Mara: The legal gap starts with SAE Level 2 vs Level 4 autonomy. Level 2 assists a human who must stay alert at the wheel; Level 4 drives completely on its own inside approved city boundaries.

Eli: Consumer Tesla Full Self-Driving is strictly contested Level 2 driver-assist. The human behind the wheel still holds total legal responsibility.

Mara: Yet that branding leaves folks wondering whose robotaxis are actually carrying passengers right now. Tesla relies on eight cameras running artificial intelligence, ditching laser sensors completely.

Eli: Here is the big surprise: those commercial cabs are not Teslas. The driverless rides roaming American cities belong to Waymo, which relies on multi-sensor suites with lidar.

Mara: Lidar sensors. As of March 2026, Waymo was handling five hundred thousand paid robotaxi rides every single week across 10 cities.

Eli: By July 2026, Waymo had driven over two hundred million fully autonomous, rider-only miles.

Mara: Meanwhile, Tesla's unsupervised robotaxis had accumulated roughly 1 million miles as of September 2026.

Eli: If Tesla isn't the company running these city fleets, can cameras alone actually pull off full driverless autonomy without lidar?

Two Eyes vs. The Lasers

Mara: That question split the entire autonomous driving industry down the middle. In 2021, Tesla made a radical engineering pivot by stripping radar and ultrasonic sensors from its cars, leaving Full Self-Driving to rely entirely on eight optical cameras. That meant rejecting the lidar and radar we mentioned, betting entirely on computer vision.

Eli: Musk insisted that the road system is designed for eyes. His logic is that humans navigate everywhere using just two biological eyes, so once neural networks solve vision, a robot car should do the same. In Tesla's view, lidar is an expensive crutch that lets developers postpone solving perception in software.

Mara: Musk compares human drivers to two cameras on a swivel neck. But virtually every other autonomous developer concluded that relying on a single sensor type creates dangerous blind spots.

Eli: They went in the exact opposite direction. Waymo outfits its sixth-generation robotaxis with twenty-nine cameras. It pairs them with six radar units to cross-check the road.

Mara: And five custom LiDAR units on top. That strategy is called sensor fusion, combining data from cameras, radar, and lidar so the vehicle doesn't rely on just one type of sensor when conditions get messy.

Eli: Software guessing distance from flat camera pixels has real physical limits. Advocates of sensor fusion argue that active LiDAR typically measures distance within two to five centimeters, while camera software can misjudge depth by meters. As Rivian pointed out, in night conditions and fog, cameras are great at seeing the world until they can't.

Mara: That puts two completely different engineering philosophies onto the pavement. How has Tesla's camera-only software actually handled real passengers on public streets without a human safety driver?

The Intervention Reality

Eli: Inside Tesla, the leadership insists the pilot ran with almost zero friction.

Mara: In July 2026, Tesla AI chief Ashok Elluswamy announced that their driverless fleet surpassed three hundred eighty thousand unsupervised miles. Those test rides spanned six cities. He claimed zero notable incidents proved cameras alone can achieve safe driving without lidar or radar.

Eli: Austin passengers had a very different experience. Videos posted by early riders showed autonomous cabs driving on the wrong side of the road, dropping riders in busy intersections, and phantom braking.

Mara: Phantom braking?

Eli: Phantom braking is when an automated vehicle suddenly slams on the brakes for an imaginary obstacle or shadow. Without a human behind the wheel to intervene, a sudden stop on a busy road quickly creates hazards for trailing traffic and draws immediate federal attention.

Mara: And regulators acted fast. On June 23, 2025, just a day after the Austin rollout, the National Highway Traffic Safety Administration said, "NHTSA is aware of the referenced incidents and is in contact with the manufacturer to gather additional information."

Eli: A single day in, and federal questions were already landing.

Mara: Then actual crashes started mounting. Reports indicated Tesla robotaxis were involved in 14 collisions in Austin after service began. Five of those crashes were reported as the most recent. Those happened across December and January.

Eli: Those wrecks resulted in property damage without severe injuries. Still, on September 3, 2026, NHTSA launched an audit query examining whether pedal-free Cybercabs actually comply with federal safety rules.

Mara: Regulators show no patience because severe failures can freeze a whole fleet. In October 2023, a Cruise robotaxi struck and dragged a pedestrian in San Francisco after another vehicle hit her.

Eli: California state officials suspended Cruise's permit immediately and revoked its license. By November 2023, Cruise had to recall all nine hundred fifty vehicles.

Mara: In December 2024, General Motors terminated Cruise robotaxi development altogether. The company had spent 10 billion dollars on the effort. CEO Mary Barra noted autonomous cabs were outside their primary business, and the capital costs were simply prohibitive.

Eli: Does bolting a Starlink satellite dish to the roof solve these fleet reliability problems and create an unbeatable tech moat?

The Dish on the Roof

Mara: Tesla certainly leaned into satellite hardware. In July 2026, the automaker confirmed on X that the fully autonomous Cybercab would feature a Starlink V5 terminal, posting, "Starlink V5 directly integrated in Cybercab." By August, Tesla shared the first photos showing that terminal fitted onto the vehicle.

Eli: The chassis itself is engineered purely for driverless trips. Tesla stated the Cybercab does away with foot pedals, side mirrors, and even a steering wheel.

Mara: Without those controls, a passenger has no way to intervene if the car falters.

Eli: Putting that dish up there ties straight back to Musk's personal commute annoyance with dropped cellular calls. Musk claimed a robotaxi needs coverage everywhere, promising Starlink for every vehicle in active markets. On an earnings call, he insisted, "We can't have robotaxis getting stuck in these Bermuda Triangles of lack of cellular connectivity," so cars do not go missing in action.

Mara: Cell dead zones are a real operational nuisance, but satellites cannot pilot an autonomous car. Technical standards reports document that satellites in low Earth orbit face frequent connection handovers, periodic signal drops, and highly variable latency.

Eli: What does that latency actually mean when a car is moving?

Mara: Latency is the time delay between sending a signal and receiving a response, much like the lag on a video call.

Eli: A frozen screen on a video chat is annoying, but waiting on a lag in traffic could be fatal.

Mara: That is why driving calculations never wait on an uplink. Electrek reported that the Cybercab's autonomous platform operates completely on its onboard computer hardware, charting paths and executing driving maneuvers without relying on any network connection. Inside Tesla reported the vehicle navigates independently off-grid, and Electrek noted that should the data link cut out, the car just keeps rolling.

Eli: So the satellite dish is just a convenience link rather than the vehicle's driving brain.

Mara: Tesla AI head Ashok Elluswamy confirmed that very distinction. He stated that satellite connectivity is "not required for safe operation of the vehicle." Instead, he explained that the link handles remote fleet management, navigation updates, and passenger customer support.

Eli: If tech features like Starlink don't guarantee autonomy, can Tesla's vertical manufacturing model still make robotaxis wildly profitable?

The True Economics of the Fleet

Mara: That internal manufacturing pipeline is the heart of the argument. Tesla reported installing annual production capacity for more than 125,000 Cybercabs at its Texas Gigafactory.

Eli: 125,000 cabs. Proponents argue building these cars for as little as $22,000 gives Tesla a massive pricing edge over competitors.

Mara: On top of that, Tesla claims long-term running costs could sink to 20 to 40 cents a mile.

Eli: But manufacturing the machine is only the start. As of 2025, commercial robotaxi running costs sat around $7 to $9 a mile, versus just $1 for personal cars.

Mara: What makes operating a fleet so much more expensive?

Eli: Deadheading. That is the miles a cab drives completely empty while waiting or traveling between paying trips. In Waymo's fleet, roughly 43 to 45 percent of all miles had no rider.

Mara: Forty-five percent rolling empty. I thought stamping out cheap cars in-house would guarantee immediate dominance, but cheap assembly cannot wipe away maintenance depots, cleaning, or remote technicians.

Eli: It really bites into revenue. An article cited an estimated yearly loss of $34,000 per Waymo One vehicle, partly driven by fares trailing Uber and Lyft during Los Angeles demand surges.

Mara: Tesla has kept its own trip margins quiet. But chief financial officer Vaibhav Taneja reported capital spending more than doubled sequentially in the second quarter of 2026, tipping free cash flow negative.

Eli: Taneja projected capital outlays topping $25 billion in 2026 to scale compute and robotaxis. Running depots and monitoring hubs burns serious cash.

Mara: Which brings us right back to our listener's question. The commercial robotaxis you hail on streets today are Waymo's sensor-fusion cars, while analysts note Tesla's vision-only system has not yet reached full unsupervised driving.

Eli: Musk admitted on a call that reliability remains the core constraint, aiming for a march of nines up to 99.999999% reliability.

Mara: Because any safety failure halts everything. As Musk told investors, "If we injure even one person, it will be worldwide headline news, and regulators will immediately clamp down on our activities."

Eli: Building cheap cars in your own factory is remarkable, but it cannot skip past the harsh overhead—or the safety proof—of running a driverless fleet.

Mara: That's it for today. I'm Mara.

Eli: And I'm Eli. Thanks for listening, and we'll catch you next time on Angle.