Krafcik Out

Leadership – Waymo

Waymo CEO John Krafcik resigned yesterday, which seemed to surprise just about everybody.

Tekedra Mawakana and Dmitri Dolgov, previously the COO and CTO, will step into co-CEO positions. As of right now, Mawakana hasn’t yet updated her LinkedIn profile, which perhaps illustrates the suddenness of the announcement.

Like lots of other folks, I have a hard time making heads or tails of this move. Krafcik joined Waymo five years ago, in the wake of Chris Urmson’s departure. The presumptive logic was that Krafcik was a business leader (albeit with engineering chops) who could take Waymo beyond research and development, and into a business launch.

Krafcik did that, technically, by overseeing the launch of Waymo One in the Phoenix suburbs last year. But Waymo One is still small and limited, and it seems like the task of launching a product has just started.

Krafcik’s stated rationale, “I’m looking forward to a refresh period, reconnecting with old friends & family, and discovering new parts of the world,” sounds an awful lot like what a poorly performing or scandal-ridden leader might say in the wake of a forced resignation.

But Krafcik is widely admired in the autonomous vehicle industry, and Waymo is currently the only company to have launched a commercial driverless business of any sort. I’m skeptical he was forced out, although anything is possible.

Perhaps it’s simply the case that the road to fully autonomous vehicles is longer than any of us expected, Krafcik spent five years getting Waymo this far, and it’s time for a break. Like he wrote.

More Waymo Challenges!

Waymo has expanded their Open Dataset to include prediction data, and they’ve posed a new set of challenges for the public to tackle.

Motion prediction challenge: Given agents’ tracks for the past 1 second on a corresponding map, predict the positions of up to 8 agents for 8 seconds into the future.

Interaction prediction challenge: Given agents’ tracks for the past 1 second on a corresponding map, predict the joint future positions of 2 interacting agents for 8 seconds into the future.

Real-time 3D detection: Given lidar range images and their associated camera images, produce a set of 3D upright boxes for the objects in the scene, with a latency requirement.

Real-time 2D detection: Given a set of camera images, produce a set of 2D boxes for the objects in the scene, with a latency requirement.

What’s more, they published an academic paper describing how they automated the labeling of the data.

Submissions to the challenges are due on May 31.

Waymo’s Safety Report

Waymo has just published a lot of information about the safety and validation of its sytems — more than I have yet reviewed. At the top level is a blog post in which Waymo breaks its safety analysi s into three parts:

  • Hardware
  • Software
  • Operations

Within each of those parts is a fair bit more detail and structure, more than I have seen in the past. For example, regarding hardware:

A vehicle equipped with the Waymo Driver has four main subsystems, which form the ‘hardware layer’. This includes the vehicle itself; the systems used for steering and driving; the sensor suite built into the vehicle; and the computational platform used to run our software.

Undergirding these descriptions are three documents:

  • Safety Report. This is 48 pages of glossy material that seems similar to material Waymo has published in the past. There’s a lot of data, but the audience seems to be more for the public and policymakers, rather than engineers and analysts.
  • Safety Methodologies and Safety Readiness Determinations. This looks neat. Lots more detail on the three layers of Waymo’s stack (hardware, software, operations). Lighter detail on how Waymo determines the safety readiness of the layers.
  • Waymo Public Road Safety Performance Data. Academic-style analysis of Waymo self-driving data from the Phoenix metro area in 2019. Unsurprisingly, the collisions recorded tend to be the fault of human drivers in other vehicles, not Waymo AVs. 
    This sentence caught my eye: “There were 47 contact events that occurred over this time period, consisting of 18 actual and 29 simulated contact events, none of which would be expected to result in severe or life-threatening injuries.”

I’m excited to read these documents over the coming days and see what they reveal. As Waymo writes in the blog post:

“There is currently no universally accepted approach for evaluating the safety of autonomous vehicles — despite the efforts of policymakers, researchers and companies building fully autonomous technologies.”

Rough Economics for Driverless Vehicles

Waymo has begun offering (“selling”) driverless rides to members of the general public in the Phoenix area. This is super-exciting, both because of the technological achievement and also because of what this advancement will make possible. And that relates to economics.

With paying customers in the vehicle, each ride covers (or at least subsidizes) its own cost. This means that, over time, Waymo can afford to drive many more miles than it could if it had to cover the cost of each ride for testing purposes.

Let’s do some math to see how that works out.

Waymo is currently logging about 1 million self-driving miles per month. Let’s assume all of these miles are part of the Waymo One service in Arizona — just for the purpose of this exercise.

Waymo’s blog post this week shared that “5–10% of our rides in 2020 were fully driverless.” To keep the numbers easy, let’s multiply 1 million by 10%, which yields 100,000 driverless miles per month.

How much money does Waymo save by going fully driverless?

Presumably the driverless vehicles are operating in low-speed environments, at least to start. That means lots of intersections and stopping. Let’s assume the driverless vehicles average 10 miles per hour. That means that, in order to log 100,000 driverless miles in a month, Waymo would have to drive for 10,000 hours.

I’m honestly uncertain of the policies and economics at Waymo One, but let’s imagine that a vehicle normally has a single operator that costs Waymo $30 per hour (wage plus taxes and benefits). That implies an operator cost of $300,000 per month (10,000 hours times $30 per hour), or $4 million per year, to log driverless miles.

That is a lot of money to me, but it doesn’t actually feel like that much money for Waymo. But now imagine scaling up.

On average, for human-driven vehicles (not Waymo), an automotive fatality occurs once every 100 million miles. Imagine that we want Waymo to drive an order of magnitude more than that, every month, in order to validate the safety of its vehicles. That’s 1 billion miles per month. Now the driver cost becomes 1 billion miles, divided by 10 miles per hour, times $30 per hour, equals $3 billion per month. That’s prohibitive, even for Waymo.

But if the vehicle becomes driverless, those costs go away.

For me, that’s one of the really exciting aspects of Waymo opening to the general public. If riders will pay enough to cover the marginal cost of the ride without a driver, then it becomes possible to massively scale testing and validation.

Volvo and Waymo Partner To Build An Electric Vehicle Platform For Ridesharing

My latest article explores the many facets and possibilities of the recently-announced partnership between Waymo and Volvo.

“Volvo and Waymo each announced that, “Waymo is the exclusive L4 partner for Volvo Car Group.” Waymo did not offer up any comparable exclusivity to Volvo. Indeed, Waymo has varying levels of partnership with Fiat-Chrysler, Jaguar, and Renault Nissan, which it mentions in the same blog post announcing the Volvo partnership.”

There’s a sensor angle, a ridehailing angle, an Uber angle, and even a China angle. Lots going on there. Check it out!

John Krafcik on Lean Production

On my way down one of those infamous web-browsing rabbit holes, I stumbled upon an article from the Fall 1988 issue of MIT’s Sloan Management Review, “Triumph of the Lean Production System,” by one John F. Krafcik.

“Really?” I thought to myself. “That John Krafcik?” How many John Krafcik’s can there be in the automotive industry?

Indeed, the article appears to be from the current CEO of Waymo, back when he was in his twenties, a graduate student at MIT.

Apparently Krafcik coined the term, “lean production.” Who knew?

The article has a lot of good stuff.

  • Krafcik’s first job out of college, before he wrote this article, was at GM’s NUMMI plant in Silicon Valley. The article kind of reads like Krafcik maybe doesn’t think so much of GM — it’s the only company he criticizes by name. (Keep in mind this is 1988, so no aspersions on present leadership.)
  • Krafcik seems to revere Henry Ford’s production system, and thinks that Japanese lean production is the natural evolution of that system.
  • Krafcik found that the location of a plant didn’t matter as much as the location of the company’s headquarters. Japenese plants in America were more efficient than American plants in America, and almost as efficient as Japanese plants in Japan.
  • Krafcik writes that European companies have a strong Not Invented Here bias that has led them to reject lean production, to their detriment.
  • Product design has a big impact on plant efficiency.
  • Plant workers should be empowered to improve processes, not just blindly follow instructions.
  • There’s not really a tradeoff between quality and productivity. High-quality plants can dispose of most inspection and rework processes, which ultimately makes them more productive.
  • Technology and robots don’t really seem to help make plants more effective.

That last point seems particularly interesting and ironic, given Krafcik’s current role.

Graph Neural Networks

A Waymo blog post caught my eye recently, “VectorNet: Predicting behavior to help the Waymo Driver make better decisions.”

The blog post describes how Waymo uses deep learning to tackle the challenging problem of predicting the future. Specifically, Waymo vehicles need to predict what everyone else on the road is going to do.

As Mercedes-Benz engineers teach in Udacity’s Self-Driving Car Engineer Nanodegree Program, approaches to this problem tend to be either model-based or data-driven.

A model-based approach relies on our knowledge (“model”) of how actors behave. A car turning left through an intersection is likely to continue turning left, rather than come to a complete stop, or reverse, or switch to a right-turn.

A data-driven approach uses machine learning to process data from real world-observations and apply the resulting model to new scenarios.

VectorNet is a data-driven approach takes relies heavily on the semantic information from its high-definition maps. Waymo converts semantic information — turn lanes, stop lines, intersections — into vectors, and then feeds those vectors into a hierarchical graph neural network.

I’m a bit out of touch with the state-of-the-art in deep learning, so I followed a link from Waymo down a rabbit hole. First I read “An Illustrated Guide to Graph Neural Networks,” by a Singaporean undergrad named Rishabh Anand.

That article led me to an hour-long lecture on GNNs by Islem Rekik at Istanbul Technical University.

It was a longer rabbit hole than I anticipated, but this talk was just right for me. It has a quick fifteen minute review of CNNs, followed by a quick fifteen minute review of graph theory. About thirty-minutes in she does a really nice job covering the fundamentals of graph neural networks and how they allow us to feed structured data from a graph into a neural network.

Now that I have a bit of an understanding of GNNs, I’ll need to pop all the way back up to the Waymo blog post and follow it to their academic paper, “VectorNet: Encoding HD Maps and Agent Dynamics from Vectorized Representation.”

The Waymo team is scheduled to present that paper at CVPR 2020 next month.

Waymo’s Full Self-Driving Experience

Recently, Waymo rolled out fully driverless vehicles to pre-approved riders living in suburban Arizona. Ed Niedermeyer has a great article (and video) in TechCrunch.

My former boss, and Voyage CEO, Oliver Cameron is a bit astounded that this event has passed with barely a ripple in the news cycle, as am I.

The lack of attention is, in some ways, a good thing.

Suburban Arizona residents haven’t gotten upset, there’s been relatively little news to make of the whole event, and so far none of the riders (who are under NDA) have found a reason to make a big deal over this.

One of questions Niedermeyer ponders is what threshold Waymo crossed that finally allowed for driverless vehicles, albeit in a tightly geofenced area.

“Waymo’s decision to put me in a fully driverless car on public roads anywhere speaks to the confidence it puts in its ‘driver,’ but the company wasn’t able to point to one specific source of that confidence….

‘Autonomous driving is complex enough not to rely on a singular metric,’ Panigrahi said.

It’s a sensible, albeit frustrating, argument, given that the most significant open question hanging over the autonomous drive space is ‘how safe is safe enough?’”

I’m not so sure I agree with Niedermeyer that the argument is “sensible”. Waymo’s response to the key question of what makes its vehicles safe enough to be driverless is, essentially, “trust us”.

And so far that works, at least for Waymo, which has done virtually everything right and caused no significant injuries, much less fatalities, in its ten years of existence.

Were Waymo to continue that trend indefinitely into the future, “trust us”, would continue to suffice.

Presumably, though, as Waymo ramps up miles and riders, collisions and injuries will happen. At that point, “trust us” probably won’t seem so sensible.

But all of that is in a hypothetical future. For now, I think it’s okay to celebrate and revel in what humanity is accomplishing.

Waymo Trucks

Jalopnik has a fun video interview with Vijay Patnaik, a product lead at Waymo, working on their trucks. I had the opportunity to meet Vijay at SXSW this year and listen to a couple of his talks on Waymo’s trucking operation. Waymo is doing great work in that area.

There’s nothing earth-shattering in the 5 minute Jalopnik video that you wouldn’t have heard from Waymo before, but Vijay does a nice job describing all of the sensors on the vehicle, and how they work together.

Let’s Talk Self-Driving

For all of the talk of regulatory hurdles decelerating autonomous vehicle development, there are important swaths of society that are highly motivated to accelerate adoption.

Waymo is joining several of these groups into a coalition called Let’s Talk Self-Driving.

“Let’s Talk Self-Driving represents a diverse set of communities coming together with the shared belief that self-driving vehicles can save lives, improve independence, and create new mobility options for all.”

Some of the prominent organizations in the coalition include the American Automobile Association, Mother’s Against Drunk Driving, and Foundation for Blind Children.

The website amalgamates lots of information that Waymo has previously shared in other formats, targeted toward reassuring communities that autonomous vehicles are safe.

Educating community groups seems like a smart regulatory strategy. Autonomous vehicles are going to improve the lives of lots of people. That will be especially true for groups that face challenges with current transportation options.

Mobilizing those groups to advocate for change is likely to be more effective than putting engineers front and center in the regulatory spotlight.