UVA’s team won the speed trial event at the Indy Autonomous Challenge last Friday.

They drove 171 mph, the fastest top speed in the events history.

By comparison, human drivers go up to 230 mph, so room to grow still.

Andrej Karpathy appeared recently on the No Priors podcast and contrasted Tesla with Waymo.

“Tesla has a software problem and Waymo has a hardware problem…Software problems are way easier to solve.”

He says that even though Waymo appears to be ahead in self-driving right now, he actually thinks Tesla’s ahead and ten years from now we’ll recognize that in hindsight.

He’s not super-explicit about what these problems are, but it sounds as though he’s primarily thinking about Waymo’s hardware problem in terms of scaling (as opposed to hardware quality, or hardware cost, for example).

That’s an interesting way to compare the two companies. I’m not sure I buy it – the binding constraint on Waymo seems more like performance and safety and regulation, rather than hardware acquisition. But you can at least imagine how that constraint could start to bind in the future.

Last weekend I took my son and his friend to a baseball game at Oracle Park, in San Francisco. As I’ve often done in the past, I parked about 2 miles away from the stadium, so that I could get free parking. Sometimes I’ve used bikes or scooters to cover those two miles, sometimes the Muni, occasionally I have walked.

This time I decided to catch a Waymo.

The ride came pretty quickly! About 5 minutes.

It parked around the corner and across the street, which was not a huge problem, but it would’ve been nice to avoid getting the kids across the street. No big deal, though.

On the way to the stadium a very nice remote support human dialed into the vehicle to inquire as to the age of the children. Only eight year-olds and above can ride in Waymos without car seats – I think that’s a DMV rule in California. We confirmed we were legal.

The route was a bit circuitous. Instead of rolling straight up Third Street, we meandered around the UCSF Mission Bay Campus. The Waymo dropped us a few blocks from the stadium, which was as close as it seemed to be able to get.

The kicker was the price. $25 for 1.7 miles. When we got to our seats, a friend who met us there and had taken a regular Uber from home – a distance of about 15 miles – told me his ride had cost $35. So it cost almost as much to Waymo the last 1.7 miles as to Uber from home.

All in all, the ride was a lot of fun. But still maybe a bit of a splurge.

We rode the Muni back to our car after the game. $2.50.

Fast Company has a nice article up about Aurora, with a particular emphasis on its proprietary first-light lidar. The sub-headling is “The Pittsburgh startup is pinning its hopes on a far-seeing lidar sensor.”

Four or five years back, a lot of AV companies were buying up sensor start-ups, although it was often hard to tell whether the acquisitions were focused on the sensor product or were purely acquihires.

Aurora has evidently stuck with its in-house FirstLight lidar sensor, and claims that sensor can see up to 1300 feet (400 meters). It’s always hard to know what to make of lidar metrics, because of the quantity vs. quality trade-off. You can configure the sensor to get a single lidar point at very long range, but at the cost of field of view, and also lots of false positive returns.

A lidar that got high-quality data at 400 m would be great. Although Aurora’s own blog post on the topic hints at the trade-offs involved.

An FMCW lidar sensor may only receive a few returns on an object 300m away, but if those hits give a velocity value of interest (e.g., moving towards the vehicle at >70 mph) our perception system knows the object is important because it is approaching quickly.

In any case, it’s interesting that while many AV companies have spun down their sensor acquisitions and gone with vendored components, Aurora is touting its in-house sensor.

The New York Times has a cool article up today about remote assistance at Zoox.

Remote assistance is mostly a quiet topic for autonomous vehicle companies. It refers to an umbrella of tools that humans in a remote operations center can use to help an autonomous vehicle.

Depending on the company, these remote assistance tools can include everything from fully driving the vehicle, to provide suggestions or information to the software stack, to bringing the vehicle to an immediate stop.

Zoox relies on guidance, but seemingly not full remote driving.

“We are not in full control of the vehicle,” said Marc Jennings, 35, a Zoox remote technician. “We are providing guidance.”

Here’s how the Times describes an encounter:

If a Zoox robot taxi encounters a construction zone it has not seen before, for instance, a technician in the command center will receive an alert — a short message in a small, colored window on the side of the technician’s computer screen. Then, using the computer mouse to draw a line across the screen, the technician can send the car a new route to follow around the construction zone.

And here’s a Zoox video on the topic.

Paul Graham, the founder of Y-Combinator, posted an essay this weekend entitled, “Founder Mode.”

Founder Mode: https://t.co/3hOnlKOJBi

— Paul Graham (@paulg) September 1, 2024

I read a lot of tweets about this essay before I actually clicked through to read the essay itself. Y-Combinator being the Silicon Valley force that it is, a lot of the tweets were sycophantic, which in turn primed me to dislike the essay.

Once I read the essay, however, I found it pretty humble and on-point, although not as earth-shattering as its promoters proclaim.

The essay starts off with a reference to a talk by Airbnb founder & CEO Brian Chesky. The essay’s praise for this talk (which as far as I know was behind closed doors and maybe unrecorded) is over-the-top.

But then the essay moves on with a lot more humility and a nice dose of concision.

Graham writes that a lot of founders have trouble scaling. They hire management teams and defer to those teams and the strategy doesn’t work out. If the founders are smart, they stop deferring to the managers, and step back into “founder mode.”

Fair enough.

The insights of the essay, I think, are two-fold. One is that non-founder executives tend be very good at “managing up,” and maybe not so good at actually managing.

C-level execs, as a class, include some of the most skillful liars in the world.

Graham’s other insight is that not much is known about this “founder mode.”

There are as far as I know no books specifically about founder mode…But now that we know what we’re looking for, we can search for it. I hope in a few years founder mode will be as well understood as manager mode. 

The first insight, about the perils of executives who are only good at managing up, strikes me as really important, although not original to Graham. I’ve worked in start-ups for most of my career and have seen and observed this. Perhaps I have even written about it. I’m confident many people noticed it before me.

That said, I think surprisingly few founders and CEOs and even start-up employees are aware of the phenomenon. If Paul Graham throws the force of his influence behind it, I think that would be to the benefit of the entire start-up ecosystem.

The second insight, about the dearth of knowledge on “founder mode” is perhaps more original and also more nebulous. Does this “founder mode” really exist? What does it entail? Graham is pretty open that he doesn’t yet know.

Perhaps, as he writes, knowing that we’re even looking for it will bring it into focus.

Uber and Cruise announced that next year Uber customers will be able to hail Cruise autonomous vehicles.

A year ago Waymo was feeling the heat from Cruise, which was launching in lots of different markets. Waymo responded by cutting its trucking business and moving faster to launch in San Francisco and Los Angeles, as well as open up rides to Phoenix Sky Harbor Airport.

A year later, Cruise has fallen behind and Waymo has surged ahead. Now Uber is feeling the heat from Waymo.

Ironically, Uber is addressing this by partnership with Cruise. The irony has levels. Alphabet (the parent of both Google and Waymo) invested in Uber, and then famously sued Uber, and now Uber is partnering with Waymo’s rival. Uber shut down its own autonomous vehicle operation due to safety incidents; similarly, safety incidents brought Cruise’s momentum to a halt. And, of course, now Uber is trying to use Cruise as protection against Waymo, whereas a year ago Cruise looked like it might surpass Waymo.

The world has come so far in a year. And in, along some dimensions, wound up in a similar place.

After a fairly long dry spell, a bunch of AV companies have raised a lot of money recently.

• Waabi – $200M
• Aurora – $483M
• Wayve – $1B
• Waymo – $5B

The numbers are staggering and reflect both the amount of money it’s taking to get self-driving technology into the market, and also the potential size of the market, which could justify these investments.

Kodiak has been super cash-efficient, having raised only about $150 million so far. We have driverless vehicles on the road for less than a billion dollars. It’s possible and we are doing it!

Kodiak recently announced publicly a project that I have been helping to lead for the better part of the year – driverless industrial trucking in the West Texas Permian Basin. We’ve partnered with Atlas Energy Solutions, a top purveyor of proppant, the sand necessary for oil fracking.

Don Burnette, our CEO, has consistently emphasized the necessity of Kodiak becoming a self-sustaining, revenue-generating business. Our industrial business, with companies like Atlas, will make that happen.

West Texas is a different environment than I-40 through urban East Texas. In many ways, the domain is simpler – slower speeds and less traffic. But there are new challenges, such as two-way bidirectional roads and intersections. Expanding the Kodiak Driver to this domain has been a lot of fun.

A few months ago I made it off the waitlist for Waymo One, and a few weeks ago I ventured up to San Francisco, solely for the purpose of taking my first driverless Waymo rides.

Years ago I had ridden around Mountain View in a driver-in Waymo Chrysler Pacifica minivan. This time, though, I used the app to hail my own Waymo Jaguar iPace rides.

A few thoughts:

• The vehicle feels very smooth at 35 mph, on wider, faster streets. My memory is that when I was at Cruise, several years ago now, we were limiting ourselves to 25 mph.
• Pickup took a while. I started at the edge of the city, took a ride to downtown, and then hailed a different ride back. On each end, pick-up took 15-20 minutes. This was late on a pretty quiet night, so presumably not much contention for vehicles.
• The “hard” in-vehicle experience is great, especially sitting in the front passenger seat.
• The “soft” in-vehicle experience is fine. Nothing bad, but nothing special, either. If you connect the Waymo One app on your phone to the vehicle, I think you can select some radio playlists.
• The rooftop dome, which has a Waymo ‘W’ icon in the photo above, is used to communicate with the public. It has some symbols for pick-up and drop-off.
• It felt a little expensive. Maybe $25 each way, when I might have guessed $10-$15. I guess Uber and Lyft have gotten more expensive, too.

Overall, it felt super normal and simultaneously a marvel of engineering. Not a particularly tough route or time of day. But just like a person.