Interview with David Silver — Head of Self-Driving Cars at Udacity
As the journey to autonomy continues, the industry will continue to require and rely on a talented workforce. While many of the world’s best computer scientists and engineers have already joined the top companies in the field, more will be needed. I spoke to David Silver, Udacity’s Head of Self-Driving Cars to learn more about how the online education pioneer is leading the way. With Sebastian Thrun as its founder, I was not surprised to hear that the company was focusing on autonomous vehicles, but I was impressed to see the depth and breadth of their curriculum. Udacity stands to be an important contributor in the industry and become one of the educational authorities on our nascent industry. My conversation with David also allowed me to reflect on the important milestones of 2017 and what we can look forward to in 2018. This is likely to be the year where we start seeing autonomous vehicles become broadly accessible to individual users. Furthermore, there is sure to be continued M&A activity that will further establish which horses will run the race. Exciting times!
Can you describe Udacity’s role within the broader self-driving-car industry?
“Udacity is online education company. Our goal is to democratize education. The link to self-driving-cars stems from Sebastian Thrun. He founded Udacity but was also the winner of the DARPA Grand Challenge that helped kick off the self-driving car boom over a decade ago. After leading the Stanford team in the Grand Challenge, he went on to form the Google Self-Driving Car team. After his time at Google, he decided to found Udacity to connect back to his interest in education. I work with Sebastian to combine these two passions.
At Udacity, we have created a 9-month nanodegree program focused on autonomous vehicles. This degree trains engineers to work in the self-driving car industry. We have had over 10,000 people enroll and currently have 600 graduates. We have thousands more that will graduate and join other graduates who have already gone on to work for companies in the industry.”
Who is the degree designed for and what is the curriculum?
“The course is a little different from others at Udacity. It has prerequisites as well as an application process. We expect applicants to have experience in programming (C++ and Python) as well as the introductory courses for an engineering program (linear algebra, calculus, statistics, and physics).
The program lasts nine months, is split into three terms, and covers 9 topics — one each month:
Term 1
- Introduction — the basics on a self-driving car
- Deep Learning — material on specific technique within artificial intelligence
- Computer Vision — a more rules-based approach to help cars detect what’s around them
Term 2
- Sensor fusion — combining output from the various sensors (LIDAR, radar, cameras) to build a holistic picture of what is around the car
- Localization — helping the vehicle determine where it is
- Control — enabling the vehicle to execute a maneuver with brakes, steering wheel, and acceleration
Term 3
- Path planning — putting together a trajectory based on perception and prediction
- Deep dive — assisting students to focus on a particular topic — often for interviews
- System integration — pulling everything together! We let student teams put their code on our self-driving car and test it.
Over the next year or two, we will be building out more introductory and advanced content. We want to bring knowledge to people who aren’t software engineers as well. Our advanced content will help go even deeper into topics we have already started covering: deep learning, path planning, etc. The goal here is to help people become experts.”
A lot has happened in 2017 — what are your predictions for 2018?
“For a long time, self-driving cars have been a research project. We’re finally seeing them become a production project that actually gets put out in the real world where customers will likely get to benefit from them. Waymo has already done this in Phoenix, where they have recently pulled the safety drivers out of some of their vehicles. It looks like we’re about to see this with a number of other companies: Uber in Pittsburgh, Cruise in San Francisco, nuTonomy in Singapore, etc. Getting ordinary people into the cars and pulling test drivers is the first big evolution I expect to see more of in 2018. The second change I expect to see is an expansion in geography. It used to be that cars were only tested in Arizona and California, places with great weather. Now, we’re seeing companies test in places with more challenging conditions. Additionally, we are likely to see companies expand their geofences. Over time, as these companies get more confident, they will open larger and larger areas, within their testing zones and beyond.
Progress is happening faster than even optimistic people expected. I wouldn’t be surprised if we see more companies open up access to SDCs to anyone in the world. Right now only Uber is doing this, with their program in Pittsburgh. Getting to the point where anyone can hail a self-driving car in various parts of the world would be really exciting. We may see this in 2018.”
With the reported reluctance and lack of familiarity with self-driving cars, how do you think companies should approach this?
“In my mind, Waymo has done a really good job with this. Their focus on working with the press has worked well. They have used press events to demystify their vehicles and getting people comfortable. They also released a safety report online that was helpful as well. Additionally, the in-car experience has helped people to understand how the car works. A lot of companies in addition to Waymo are focusing on user experience to get people comfortable — I think this will approach will be successful.
The bigger question is if we start to see the cars getting in accidents. People may start to lose trust and even fear self-driving cars. I am not quite as worried about this since human drivers are themselves quite dangerous. That being said, it won’t take much for SDCs to spook people if we don’t get it right. There is a responsibility that self-driving companies have to the industry and to society not push out their product too early. Inevitably, however, there will be some accidents. We need to educate the public about the real issue at hand: human drivers are not safe and automated driving will be much safer.”
Where do you see the industry trending with regards to a proprietary vs. an open-source approach?
“It will be fascinating to watch this play out. The comparison many people use is the mobile phone industry. In this ecosystem, we wound up with two big players. First, we have Apple’s iOS, which is proprietary and closed and has a loyal following. Second, we have Google’s Android, which is largely open-sourced and has a wider distribution. The latter may have broader reach, but handset manufacturers are making less money than Apple, and even Google is probably only making money indirectly.
With SDCs, it’s not clear how that the business model will work out. It’s interesting to see Waymo take more of the Apple approach. The OEMs have seen how the phone industry played out and do not want the same thing to happen to them, where another player captures the majority of the value. There is a widespread belief that the data and machine learning models are where the value lies.
Another interesting element is that of network effects. With mobile phones, apps drove network effects to a great extent. It’s not obvious yet if network effects will be as important in autonomous vehicles. It is possible: having more data and being open-sourced could be beneficial in developing and maintaining the system. But if that turns out to be less imperative, then we might see lots of different companies working on their own.”
Who should we be paying attention to that we aren’t?
“Self-driving cars are going to change the world in ways we can’t even imagine. Therefore, there is a straightforward answer and a more complex answer here. There are a ton of automotive companies that are slightly less known but are going to be very involved in this revolution. The Tier 1 automotive suppliers (Bosch, Delphi, Continental, etc) may see their role change completely — going from being suppliers of the big car manufacturers to being competitors. Look at Delphi having acquired nuTonomy — that could be a really interesting business model change for them.
A similar thing can be said about Tier 2 automotive suppliers (Nvidia, Velodyne, etc) — they have the potential to be transformed because the demand for their products stands to be huge.
In a broader sense, if we again look at the mobile phone industry, the biggest company to be born out of this wasn’t directly in the industry. It was Uber. Smartphones allowed Uber to disrupt the transportation industry during the mobile phone revolution. As mobile phones were becoming more and more widespread, I spent a lot of time thinking about what would be disrupted. I could see how commerce, banking, and dating would change. I never thought about transportation and that’s where the biggest disruption of all took place.
For AVs, we don’t know what will happen. The same thing could take place again in a different industry — many people can see the upcoming changes in real estate, hotels, food, or childcare. But there might be things like medical care or another unknown that will be completely transformed.”
What do you expect of the big tech companies such as Facebook, Amazon, and Microsoft?
“As far as self-driving cars, Facebook seems less concerned and is more active on Virtual Reality. Their focus on Oculus could end up being extremely profitable for them. Amazon has a robotics program and is obviously focused on logistics and delivery. They stand to gain a lot from the changes in transportation. They have an obsession with lowering cost and will likely be involved in the future. It’s unclear how much they will do themselves vs. relying on others. Microsoft has historically done quite a bit within the vehicle’s infotainment system, but less on the vehicle hardware and software itself. They have focused on the connected car and are slowly starting to make their way into the automotive space more broadly but still on the periphery. They have yet to commit to self-driving cars in big way.”
What is the issue in this industry that people aren’t focusing on but should be?
“I can think of two in particular. First, people don’t realize how costly the maps are that enable these vehicles to be autonomous. This may seem boring but it has dramatic effects on how the vehicles will get rolled out. The amount of data that the sensors take in makes it hard for them to rely only on real-time sensor data. They need more and the high definition maps have so far been the answer. However, they are incredibly expensive to produce and get outdated all the time with changes in the roads.
The other interesting topic is the debate between going directly to autonomous vs. a more incremental approach. Tesla is taking a very different approach from the rest of the industry. They are trying to layer on more and more driver assistance features to eventually get to full autonomy. On the other hand, companies like Waymo are attempting to go directly to level 4 but have to do so in tightly geofenced areas with high definition maps. This is very expensive. Waymo and others are hoping to have the computer handle all the driving immediately. Tesla is taking the approach where the car has to be able to be driven by the human and a computer. The race is on to see who will get there first.”
