Tech Brew Ride Home - (Bonus) Waymo's Head of Hardware, Satish Jeyachandran

Episode Date: October 5, 2019

Satish Jeyachandran is the Head of Hardware at Waymo. Before that he lead the self-driving team at Tesla, so he was in charge of the Autopilot team, and his job change lead to headlines at the time. Today Satish gives us a history lesson on Waymo, tells us what the future holds for self-driving tech generally, and most importantly for me, answers a question I’ve always wondered. Why go for full autonomy? If you can give me the ability to let the car drive itself on highways, why not give me that now? Why try to solve the full 99.99% of the problem? It turns out, that Waymo learned you can’t do half measures. You have to do full autonomy or you do nothing. Very interesting conversation. Learn more about your ad choices. Visit megaphone.fm/adchoices

On April 4th, 2023, around 2 in the morning, a man was found stabbed multiple times on a sidewalk in downtown San Francisco. Hey, who did this to you? What happened next turned the story into a political firestorm. Reports have identified the victim as Bob Lee, the founder of Cash App. From Bloomberg Podcasts, this is Foundering, the Killing of Bob Lee, beginning April 16. Welcome to a weekend bonus episode of the Techmeme Right Home. I'm Brian McCullough. Satish Jachandrin is the head of hardware at Waymo. Before that, he led the self-driving team at Tesla, so he was in charge of autopilot.

And his job change from Tesla to Waymo made headlines at the time. Today, Satish gives us a history lesson on Waymo. Tells us what the future holds for self-driving tech generally, and most importantly for me, answers a question I've always had. Why go full autonomy? Why go the full bore? If you can give me the ability to let the car drive itself on highways, then why not give that to me now? Why try to solve 100% of the problem if you can already solve 95% of it? Well, it turns out that Waymo learned early on that you can't do autonomy in half measures. You have to do full autonomy or you do nothing. Very interesting conversation. Thank you, Satish. Let's start with sort of a history lesson because I feel like, you know, to the degree that you can be a professor on this, I feel like the history of Waymo is not very well known. And you don't have to go super deep on this. But can you, just in broad strokes, tell me how the project that became Waymo began? It started in the X lab, right, around 2009?

Project has a roots in the DARPA Grand Challenge. days. A lot of our core members did come from there. So it was the ideation was inspiration from that project on and it started in 2009, 2010 timeframe and we have

progressed since then. We were part of Google back then. Then we moved on to the Google X umbrella. Then we have graduated since in 2017 to Waymo. So it's been a progression

for a while. So for the first seven, like eight, ten years, we've been part of the Google umbrella. Very early on, the Waymo, now Waymo, but the project, you guys started designing its own sensors in-house, I think around like 2011. And my understanding of that is that it was basically because there wasn't anything on the market that could deliver the capability. that was felt was necessary for a car to be fully driverless? Is that right? That's very true. Yeah, we started the government back in-house in 2011.

Before that, we did look at the market pretty widely and see what's available in the market, and we did try a lot of these sensors before deciding to do the work in-house. Pretty much the industry was focused on getting the safety features out, like emergency braking kind of features out, and most of the sensors were tuned to it. But we had pivoted to focus on L4 capabilities, and there was a huge gap between what the industry was offering us and what our needs were.

Was that? That was the main reason why in back... Sorry, go ahead. Go ahead, Ryan. That was the main reason back in 2011. We brought it in-house here. Well, so what I was going to say is it was that sort of, you guys had greater ambitions than anybody at the time,

where, again, like, it's automatic braking, and it's maybe like, you know, advanced cruise control or something, but it's basically maybe at the time the team was thinking more ambitious than anybody else in the market was. Yeah, our goal so much were wider. We really wanted to make the roads really safe. And it was, yes, you're absolutely right. We are much more ambitious project than where the industry was here. I believe that Waymo basically designs the entire sensor suite, like the hardware, the software, the computing, the full stack. It's all in-house to this day.

Is that true? And is that because your technology is better or no one's caught up? Again, like if it's, you know, back in the day, they couldn't take anything off the show. is it because no one's been able to basically provide you sort of those off-the-shelf tools that would be useful? Yeah, I think over the course of the years, we've driven close to 10 million miles, and over these miles, we've learned a lot of requirements and experienced a lot of user scenarios that has kept on fueling our improvements on the sensors and compute and so on. And that has primarily fueled our iterations over the years.

And the second phase is like the industry as, like I said, nobody has driven as many miles as we have done. The industry doesn't have the full know-how of what is really needed to do these kind of work. And there's still some areas of gaps between the industry and what we do at Waymo. We are always at the forefront of what is needed going on. So for example, like our shot-range LIDAR, when we, put it in the market and we didn't have a product which was able to look 360 degrees and and had a 95 degree voltage field of you that can do take anything next to your tire all the way up. We didn't have it. So we had to go to the drawing board and come up with a sensor like this.

Another example is our radar when when the industry is still focusing highway driving, we added to develop a custom high sense to 360 degree data to work in urban environments when it's challenging when you go through a huge building. things, tunnels, clutter on the road and all the stuff. So our radars are specifically tuned to work in the urban space. So those are some examples where we have pushed the technology to suit the L4 needs and urban driving. You've already sort of described it a bit, but again, this is sort of like a deeper dive show where we're talking about, you know, autonomous vehicles all the time on the show. But, you know, and we're using terms like LIDAR and things like that.

But can you describe for me, like, not the secret sauce, but the overall, like, what Waymo's technological package is? Like the LIDAR, the vision system, the radar system, like, how do they all tie together? So, yeah, three pillars of our sensing system are the LIDAR, the radars on the camera. The LIDARs, I think we have three types of LIDARs in the system today. One is a medium-range radar, a LIDAR, which is 360 degree, and a short-range LIDAR system, which are many of them are on the car to give close and sensing. And we have a long-range LIDAR, which provides very long-range detection at very far distances, so where we can see certain stuff on the road much earlier than what a medium,

range of light arc could do. The combination of these three give an unparalleled 360 degree field of you both close to the car and very far away from the car so we're able to paint pretty much a picture around the car which helps our upper level software to detect certain things very reliably and also provide velocity and range to that objects. So that's why our LiDAR system is very unique because we're able to merge all three systems to give a full 360 degree field of view. Similarly on the radar system, we do have multiple radars on the car and they are seamlessly able to provide a full 360 degree sensing of the real bolt. So similarly a cameras have high dynamic range and the resolving power is much greater than what we have what we have seen in the industry and they're

able to provide the focus in all temperature ranges. So when we combine all three main pillars, the perception system on the air on top of our hardware is able to get a a full 360 degree view of what's around a car, and it's getting information from multiple domains. Modalities, what if I can say that, because it's coming from the LiDAR space, it's getting from the RF space and radar, it's getting it from the visual space and the camera. So that uniqueness of using the data is like pretty awesome for the upper level software. The election ride home is a daily podcast devoted just to the 2020 elections. Every day at 5 p.m. journalist and this American life contributor Chris Higgins brings you the latest news from the campaign trail.

Who's up, who's down, what issues are getting traction, and what do the polls say? Someone is going to challenge Donald Trump for the White House, and every day in 15 jam-packed minutes, the election ride home helps track who that person might end up being. Search your podcast app and subscribe to the election ride home podcast. So that is, that sort of holistic thing is like, it's, it's, it's, it's a lot of, it's, is what makes your system different than what others do? Like, are you just more the whole package? Or, like, what differentiates what Waymo system does than what other systems are trying to do?

Yeah, the technology, definitely there's a differentiation in technology. We are in our third generation's the hardware suite, which is custom developed for L4 systems. So we have a lot of learnings from the previous generation. And the uniqueness to Waymo is, like, we have hardware engineers and software engineers working together. in the same space and there's no siloing between them so we're able to get real-time feedback what's working what's not working we also have our own data

accumulation and operations where we get a lot of information from what scenarios happen in the world so this know-how of finding the requirements what is really needed for L4 and driving the technology through multiple generations of development has kept way more head of everyone and we'll keep on going ahead as we try more miles and we expand and so on. So those are some of the advantages of Vamo. I would say technology, the people, collaboration between our hardware and software, our experience over the years.

I'm going to come back to sort of like your secret sauce in a second. But like my general understanding is that it's sort of like when we come to think of autonomous systems, like, you can be 95% of the way there, you can be 98% of the way there. And actually, this is what we're going to get into now, this idea of, like, you know, safety and, and, and things like that. But my understanding is, is that, like, it's sort of like, you know, a hockey stick sort of graph. Like, so that last 1% or that last 1 tenth of 1% or 1,100th of 1% in terms of, like, what the systems are capable of, it gets progressively hard. harder and harder. Is that, am I off on that estimation of like what the challenge is for your system? We are not different from any other technology curve. As you get closer and closer to the

market or the customer acceptance, it gets challenging and challenging. So we are not different from that aspect. But we have a unique advantage that we have more than a billion miles in simulation and 10 million miles in driving. So that adds more. information and increases our confidence towards that. Let me, I don't know if you're familiar with this anecdote that I found researching before we spoke that your CEO shared at IAA Frankfurt recently. It's a story about how in 2013 Google ran an experiment where they gave semi-autonomous technology to a handful of employees that had long highway commutes.

Do you know this story? Yeah. Yes. Yes, so basically they say this is early on, this is before you become Waymo, and they give the technology to the employees and they say pay attention, keep your eyes on the road at all times. You can take your hands off the wheel, but you got to be alert. So then finish the story for me.

Tell me what happened when that happened. We said, hey, if we keep the human in the loop, we still at a position where they take it as we progress the technology, Until we're 100% sure, we cannot bring the human in the loop. Because what happened... What happened was is they were still being watched by cameras, and they didn't follow the rules. Essentially, they were putting on makeup, texting, falling asleep even.

So as you're saying, that was a huge lesson to the project? So that's when we pivoted from driver's asset kind of feature to like L4's. driving. That's the pivot because human in the loop, people start to be realized people started to build confidence after so many hours or miles in their car. They start a build confidence that they get distracted and they do other stuff. Even though we clearly told them you need to have your hands on the steering and focus on the road. But it showed us a weakness in the human behavior where people after people, different people have different conference level, but it's easy to get your confidence of humans.

Okay, and you describe that as a pivot. That's why I was so glad that I found this anecdote because this is something that, again, as a layperson in this field, like, I'm always like, well, they want to get, you know, driving in Manhattan right, they want to get driving in snowstorms, right? But if they can give me now the ability to be on the highway and have like some super cruise control or something, well, why can't they just give that to me now?

But what you're saying is, is that by running that experiment, Waymo, it was before it was Waymo, but the project had to pivot because they knew that humans instinct, humans would naturally trust the technology more than they could. So it was almost like you can't do level two. You have to go all the way to level four or nothing. Am I being too grandiose in describing it that way? You're exactly right. Yeah.

That was our high-level takeaway. So that essentially you have to go all the. way or nothing because any half measures will will be unsafe and and and won't work you have to go all the way or nothing yeah you put it right away the scenario in the world is different right different cities different weather conditions and people behave differently so you cannot predict everything you need to build a system which to handle everything by yourself or or you might risk the human behavior You cannot say I'm good at 98% and 2%

you need to always pay attention to get in. So, yeah, you well summarized it. How's Waymo 1 doing? I'm not asking for any, like, updates or press releases or anything, but are you pleased with the progress of that project? Yeah, it's been a great beginning for us. It's a good learning experience for Waymo and the team. And the customers also have been very enthusiastic,

I mean, we have a crowd where they're so passionate about this technology and making the road safer. We're getting very valuable and very critical feedback, and so far has been a very great start. Is there any new technology coming down the pike that you're excited about, like the future of sensing for self-driving beyond things that, again, we've talked about, like, LIDAR and radar and the vision systems? Is there something new that might, like, push things to the next level? I think trends within the same LiDAR, radar, camera, speeds.

The technology is slightly deviating from one another. So there are some pros and cons of those. We have constant evaluation and playing around with these technologies. Also, from Waymo side, we focus a lot in expanding throughout the world. So we focus a lot on weather condition, different driving conditions, like you mentioned, New York. Driving in New York in certain months or weeks, it's challenging. So our focus has been driving the technology towards

making a sense as more reliable and useful in those conditions. So that's how I've been our future kind of stuff. Keep an eye on what the industry is doing and what the emerging technologies are, push forward our expansion.