The day has come: it’s time for the reveal of the Tesla Robotaxi (er, CyberCab?) After years of Tesla CEO Elon Musk’s promises to solve self-driving, the automaker is poised to reveal the culmination of its efforts. And with it comes a lot of questions. When will it launch? In what cities? Can I use my Hardware 3-equipped Model 3 like a taxi as promised when I bought it? Can Tesla's AI truly be trusted to take the wheel?
Welcome back to Critical Materials, your daily roundup for all things EV and automotive tech. Today, we're chatting about the Robotaxi's AI problem, the auto industry's battle for and against LIDAR, and BYD breaking into Mexico. Let's jump in.
30%: Tesla's AI Has A Black Box Problem
Today, Tesla will reveal its futuristic, driverless, urban mobility mobile that could make ride-hailing cheaper per mile than owning a car. That’s a tall order, I know. And whether or not Tesla lives up to that task is another story.
What we do know is that we still have an underlying problem with our robotaxi overlords, and that’s Artificial Intelligence.
According to a number of industry executives, autonomous vehicle experts, and even one Tesla engineer who spoke to Reuters, the thing that's supposed to be helping autonomy is actually one of the major weaknesses of Tesla's approach.
The weakness that the engineer is describing is an inherent phenomenon of AI: the "black box" problem. But before we get into that, let's understand how AI learns.
There are two big pieces of the AI puzzle—training and inference. When a model is trained, gobs of curated data are thrown at it to teach the model how to make decisions. How to approach stopped traffic, how to recognize a red light and how to safely navigate an unprotected left turn. These are all things that a new driver needs to learn to do, too.
The problem is that training takes a ton of power and resources, like computing power and storage. It's the reason that Tesla has had to build giant multi-billion-dollar data centers dedicated only to training its self-driving model. It's not feasible to deploy that same level of hardware to a car.
That's where inference comes into play. Inference makes decisions on how to infer the real-world data around the car based on the trained model that it's fed.
As pointed out by the engineer and industry experts talking to Reuters, the "black box" sensation is the lack of understanding of why end-to-end AI—that's the ability to feed a model completely raw data and it produces decisions without an interim engineering or programming steps—makes the decisions that it does.
"[It's] nearly impossible [to] see what went wrong when it misbehaves and causes an accident," said the Tesla engineer in a statement to Reuters.
The engineer continued to note that it's not the failures themselves that are necessarily the worry, but the inability to safeguard against those types of failures in the future. And that leads to a lack of accountability and transparency when attempting to verify not if the vehicle performed a particular action autonomously, but why the vehicle chose to take that action specifically.
It's not just Tesla that's worried about the black box problem. Jensen Huang, founder and CEO of Nvidia (which supplies a massive amount of the processing power behind Tesla's newest data center in the form of its H100 GPUs and is also working on its own autonomous driving system) has also brought up concerns about not being able to understand how end-to-end AI makes its decisions. Despite not being able to understand it, this method typically, but not always, results in the "best" driving decisions.
"We have to build the future step-by-step," said Huang. "We cannot go directly to the future. It's too unsafe."
Today's robotaxi unveiling will undoubtedly involve some flashy new tech, ambitious promises, and a timeline that will likely be stretched with age. But once the show is over, the real work behind the scenes will begin. That hard work will be necessary for humans to feel a bit more trusting when taking a trip in one of Tesla's robotaxi on a real road with other human drivers around occupying the streets.
60%: The Auto Industry Can't Decide If It Likes LIDAR Or Not
LIDAR is either a godsend or a crutch, depending on who you talk to. And now, years into the self-driving race, it seems like the auto industry still can't make up its mind. Should we spin our wheels and just hope that cameras are enough to accelerate cars to the Level 5 autonomous dream that some hope for? Or is that time so far away that we need LIDAR to get there quicker?
That's the debate that the industry is facing.
LIDAR, or "Light Detection and Ranging," is a fancy piece of technology that helps devices visualize the world. It's the reason that your iPhone camera is (usually) good at focusing on a subject, and how certain autonomous vehicles—like those operated by Cruise and Waymo—turn everyday objects into a point cloud of ones and zeros. For lack of better words, LIDAR sensors are the eyes behind the software.
The problem is that the industry isn't so hot for LIDAR anymore. It used to be, though. But in the last year, LIDAR providers like Innoviz, Luminar, and others have seen stock prices dip as much as 75%. Automakers and investors seem to be losing their faith that the tech is the secret sauce in solving self-driving.
"At a much earlier stage, there was an appreciation of the long term, and arguably, there was too much of everyone believing anything and 'everybody's a winner,' " said Luminar Technologies's CEO and founder, Austin Russell, in an interview with Automotive News. "Now, it's flipped to the complete opposite and extreme skepticism."
Most automakers aren't buying into the whole "throw a LIDAR unit on every car" idea, as much as Luminar would love that. The reason? Well, LIDAR is expensive. As in, tens of thousands of dollars expensive up until a few years ago; now, Luminar sells units for around $1,000. So throwing LIDAR sensors on every single car rather than a suite of cameras and other radar sensors that might one day be able to do the job sounds like a sound financial plan until automakers can figure out the software piece of the puzzle. After all, if LIDAR users like Waymo and Cruise haven't figured it out, it seems like a waste of money in the short term.
Then there's Tesla, which, despite being Luminar's biggest customer, believes LIDAR is a straight-up waste of money in its cars. Tesla believes that it can solve the autonomy problem with cameras alone—as in, no long-range radar, no ultrasonic sensors. Just cameras. It's been promising that for nearly a decade.
But then again, Waymo has been developing its own LIDAR sensors for even longer than that. And if Waymo hasn't been able to solve the problem flawlessly yet (and cameras are providing a "good enough" solution for most drivers) maybe manufacturers just don't the value in investing in LIDAR yet.
So here's where we come to an impasse: money, technology, and time. The industry has developed a weird "choose two" triangle while treating the tech like Blu-Ray versus HD DVD. And the only way a winner will be decided is by one of them solving self-driving first, for a palatable price.
90%: BYD Wants To Sell 100,000 Cars In Mexico Next Year
BYD, king of the cheap Chinese EV, isn't messing around. It's been devouring the competition back home and has been silently expanding its footprint to feed its insatiable appetite for market share. Mexico is next on the menu.
We've known that BYD has had its eyes set on Mexico for some time. The automaker has been looking at spinning up a new EV plant there for a few years, though recent rumors suggested that the company would wait until after the U.S. presidential election to announce if (and where) it would build the plant. As it turns out, the plant is happening, and BYD has committed to announcing the final location by the end of the year.
BYD has some rather ambitious plans for this plant. In fact, it expects to use its new assembly facility as a weapon in its goal to sell 100,000 vehicles in Mexico by 2025—that's double the figure it expects to sell in 2024.
Jorge Vallejo, BYD's head honcho in Mexico, solidified opportunistic production numbers, noting that the factory will produce two separate production phases, each of which set to produce 150,000 vehicles—though a timeline for production, or explanation of what these phases mean for foreign markets, was not given.
Here's the thing: BYD has said it has no plans to expand into the U.S. market at this time. And the adoption of recent protectionist tariffs by the U.S. makes the idea of an affordable EV by a Chinese automaker more of a pipe dream than a reality. Heck, even Canada has followed in the same footsteps as it too announced hefty duty fees just as BYD began eyeing up an expansion into the Great White North. So where does that leave BYD in North America?
One thing BYD has going for it is a rock-solid supply chain. The automaker has proven itself to be exceptionally good at making quirky EVs that people actually want to buy. Whether that's because of specs and features or cost is another story, especially when its uber-cheap Dolphin retails for right around $16,000 back home. But BYD is able to build them quite easily and rapidly, sending its position as a global EV powerhouse skyrocketing.
With BYD plotting a plant on America's doorstep, it might be able to cut costs down enough to do battle with America's legacy automakers—even with a 100% tariff artificially inflating the cost of its cars. Because if there's one thing that we Americans love, it's consumerism. Give us a good car, and we'll buy it (probably). If BYD can figure out how to sell it stateside, then they may have a winning recipe on their plates.
100%: It's Robotaxi Day! What's Gonna Happen?
Well, folks, it's that time: it's Robotaxi Day. Or the CyberCab Symposium. Or We, Robot. Whatever Tesla is calling it, it's time for the automaker to unveil something that will either make or break the company's big bet on autonomy.
I want to hear some predictions here: what's going to happen tonight? Will Tesla show off some hyper-intelligent stock pumper? Or will it be another person dancing around in a spandex robot suit? Let me know in the comments.
Contact the author: rob.stumpf@insideevs.com
