American Autonomy: How Lucid and Rivian Plan to Deliver What Tesla Just Keeps Promising

• Tesla has been promising unsupervised self-driving capabilities for more than a decade.
• Lucid and Rivian are letting their hardware and software do the talking.
• How companies approach development and safety should matter to consumers.

But while Musk continues to make more promises, two American rivals have been quietly working on the self-driving problem. Unlike Tesla, Lucid and Rivian are taking a build-first, hype-later approach.

This article looks at what it takes to actually deliver autonomous driving safely and at scale, focusing on the progress of automaker-led plans, specifically those of what I like to call America's "E-Three" EV companies: Lucid, Rivian and Tesla.

Perception: Sensing the world

Mercedes-Benz was the first and so far only automaker to offer a Level 3 (eyes-off) autonomous vehicle to customers in the U.S. The option was offered on select vehicles sold in California and Nevada, the only states where the car was allowed to take over. (Patchwork autonomy laws across different states are a different issue for another time.)

The Mercedes Drive Pilot system relies on high-definition maps captured using lidar — among a host of other sensors including cameras, radar and ultrasonics — to sense the surroundings and determine the vehicle's precise location. Owners couldn't get very far, and neither did this program. Mercedes-Benz's next shot at autonomy has already been previewed on the new CLA, and it's much simpler, a Level 2 (eyes-on) system that does without pricey lidar sensors or the high-def maps their output is checked against.

Tesla famously wrote off lidar years ago and even removed the radar and ultrasonic sensors from its cars in favor of a camera-only sensor suite. Despite this choice having some very obvious downsides that have led to well-documented accidents and glitches — cameras don't do well staring at the sun and can't reliably perceive depth, speed or position as well as other modalities — Musk and Tesla continue to stand by it, both for customer vehicles and company-owned robotaxis.

If you're really and truly going to take the driver out of the vehicle, most companies want as much sensor coverage as possible, both with overlapping fields of view and multiple modalities, which complement each other. 

Waymo is one such example on the commercial side. The latest version of the Waymo Driver sensor suite includes 13 cameras, four lidar, six radar and some exterior microphones (Waymo calls them EARs, or external audio receivers). Motional, which is backed in part by Hyundai, uses a similar setup. Note that these services are currently geofenced and rely on detailed custom maps of the roads on which they're allowed to drive. They also employ so-called sensor halos, large roof-mounted collections of gear giving them 360-degree awareness, hardware that wouldn't fly for a consumer product for both aesthetic and cost reasons.

Lucid and Rivian baked in a variety of sensors on their first products, the Air and R1S/R1T, respectively. All use short- and long-range radar, ultrasonics and a host of cameras, and Lucid adds lidar on vehicles equipped with the DreamDrive Pro package. For its second act, the R2, Rivian will offer an available lidar sensor that's neatly integrated at the top of the windshield. Even without lidar, Rivian has given the R1 Tesla-like capabilities. Lucid added more sensors on the larger Gravity, the SUV that followed the Air.

Like Tesla, both companies launched with the hardware and added more driver assistance and hands-free driving capabilities as they were ready through over-the-air updates. Unlike Tesla, these companies underpromised and overdelivered in terms of features and timing.

Redundancy: Backups with backups

Sensors are no good if they lose power or otherwise fail. Redundancy of critical systems matters, too.

In its Level 3 system, Mercedes-Benz uses redundant braking and steering actuators, separate power sources and duplicated computing — basically two CPUs performing calculations simultaneously in the event that one goes down. With its latest MB.Drive Assist Pro, there are redundancies in the software stack, but the driver is the ultimate failsafe, which is the norm for Level 2 systems.

Because Tesla is Tesla, it doesn't share a lot of information about what's underneath the skin of its vehicles. Independent reporting and teardowns have turned up some information about redundant systems. The full steer-by-wire system used on the Cybertruck features redundant power and controls, with two motors incorporated into the steering rack, likely fed by separate power sources. It's a good bet that a similar system is used on the controls-free Cybercab that Tesla plans to deploy as part of its Robotaxi service. Note that redundancy like this is not present on the Models S, X, Y or 3 that came before the Cybertruck. And without such backups, many of the Full Self-Driving promises simply could not have been fulfilled safely. And with Tesla, sensing redundancy is achieved by adding more cameras. But that's no help if none of the cameras can interpret the scene correctly.

Before it even adds a Level 3 system, Lucid has backups in place on the Gravity. Its zonal electrical and communications architecture allows different portions of the car to be powered by different supplies. The vehicle is designed so that if one power source goes down, the other portion has enough sensor coverage to safely pull the vehicle over without driver intervention and even keep one headlight on.

All Rivian has said about redundancy on the R2 is that its new compute architecture does the mirrored, simultaneous calculations described by the ISO 26262 standard, which outlines functional safety as it relates to automotive systems.

Compute: Custom chips and necessary upgrades

Tesla has iterated on its in-house chips for years, with the company continuously promising that the next version is the one that will get its vehicles to actually drive themselves without supervision. This promise-first routine has gotten Tesla in some trouble with owners, some of whom were originally told the cars they were buying would one day be autonomous (Full Self-Driving in Tesla-speak) and are still waiting for the promised feature.

Here again, Tesla has been promising capabilities that it has to admit its cars can't handle without swapping in an upgraded computer. And without hardware redundancy, true Full Self-Driving wasn't even a pipe dream.

Rivian and Lucid haven't been around for as long, but they've both waited until there was something to say about autonomy before saying it. At its recent Autonomy & AI Day, Rivian announced plans to build its own processors for future models, part of what will enable the addition of lidar for the R2 this year; the R1 platform currently uses Nvidia hardware. Lucid is sticking with Nvidia for the time being, both for its consumer vehicles and the Gravity robotaxis it is developing with Nuro and Uber. Those companies didn't announce their partnership until after the vehicles started testing.

All this is to say that the order of operations is important. Develop the technology, then tell the public it's coming and when. Selling a car based on its potential future capabilities is at the very least disingenuous. If consumers are going to trust the autonomous future, they need to be presented with the facts of the autonomous present.