Toward a Driverless Future | TC Sessions: Robotics + AI 2020

Foreign [Music] Gentlemen thank you so much for joining Us thank you we're here uh you said Backstage Jordan said we're here to talk About cars but you said trucks that's Right yeah your company does trucks and Then Aurora of course does cars we do Cars and trucks yes yeah it's really Interesting because both of your Companies are addressing autonomous Vehicles but different markets but at The same time you both have one concern Which is safety it was kind of funny Because two years ago this month An Uber self-driving car struck and Killed a pedestrian what has changed in The market since then I don't know that That I would say The like the case for safety has changed Like it's always been pretty absolute Right you're you're trying to focus on Building a driverless platform that you Can produce a a functional safety case With a Strong pedigree and you you Demonstrate that internally to your own Processes and then you demonstrate that Uh to uh regulatory satisfaction and and Then ultimately it is the public that we That we need to satisfy that we are Building safe vehicles that they can They can trust operate around them Obviously

Um you're not with Uber but how did that Incident change Awards Outlook and Safety uh I can't really say because I Like I was not at Aurora at the time um And uh but Aurora has always been Founded uh centrally Um with a a a a core belief in safety Like the the mission is to remove the Driver and in order to do that Um a lot of the challenge comes around Not just how you build a system that Performs well but how you actually Validate that that system is safe in its Performance Um and so you know Aurora Aurora's Mission statement is to deliver the Benefits of self-driving safely quickly And broadly Um and uh and they you know they were Very careful about choosing the ordering Of those of those statements that safety Will always come first and that we are Willing to take more time we are willing To delay the the emergence of true Self-driving vehicles to really build a Safe and validated platform yeah so to Go back to your first question I was Like what has changed in the market I Have seen a think a shift in the space In terms of how to uh approach the the Technology so when we founded the ike it Was not like how can we get a prototype On Road as quickly as possible but like What does it actually take to safely

Deploy a product on on the road sure What does it take to to build that and Let's build that so that that kind of Shift as you've seen broader I think and Or as an example of that as well uh Where where there is a just a different Approach to developing the technology Um and then I think that is partly Um yeah motivated or reinforced by what Happened at Uber at the time right so Ike does long distance Trucking correct Where does pedestrian safety or other Drivers safety come into play with with Your company yeah and that's of course An amazingly important because we are Doing trucks on the highway or core Domain so we are not envisioning Transporting passengers ourselves so the Safety case is always with respect to The other participants of traffic uh so Yeah that's from the center obviously Like they need to be safe both of your Companies rely heavily on simulation for Safety testing what points do you look For to say okay it's safe now in the Simulator we can take this on the public Road what do you look for So um That our our kind of virtual testing is Is largely uh The safety is is one component but it it Kind of underpins the entire program a Lot of that is around how we develop Quickly how we are able to kind of

Explore the space of solutions and um And so We look at it as uh it's very important To move quickly and to like to try a lot Of different things and to empower Engineers to run a lot of different Experiments Um and you're always being slightly Constrained if if the only environment You can test that on is is is on public Roads or or even even in in private uh Tracks you you still have to like you Have to worry about certain certain Limitations of your solution and so with With virtual testing you you kind of Open up the exploration space to to try A lot of different things and you can Also Um you get a lot of feedback very Quickly Um about Corner cases and edge cases Um and so a lot of what we do is really Allowing uh rapid iteration and Rapid Feedback and then um and then doing that In a way that provides us confidence That that we shouldn't have any uh new Findings on road and where we do have New findings on road when we when Something you know slightly unexpected Happens or you know we'd have a Disengagement because uh we would take The vehicle operator the safety operator Would take over we always roll that back In right so into into our virtual

Testing so uh I mean this model is Fairly common in lots of Industries Where when you when you don't have the Ability to do full full-scale testing is That you are building Um you build a a virtual environment you Build a simulated environment and then You are constantly using real world Learnings either a subscale subsystem Testing or full-scale testing and Rolling back back into the Fidelity of Simulation what about at Ike yeah so of Course we all know that if something Does not work on simulation it will for Sure also not work in the real world if It does work in simulation it may work In the real world right so we use Simulation like throughout the the Software and Hardware in the loop as Well to test uh all of our development Uh and quickly iterate it's basically The same same idea across all components Of the system and only when and we have Like work through the simulation that The simulation works really well we may Want to test it on a public uh not on a Private test track we have not gone on Public roads because for the development Of our technology uh that is not Necessary and uh and it incurs a lot of Risk and uh and uh and and cost to do That so we we really focus heavily on Simulation and we have a private test What platforms do you use for simulation

Are they off-the-shelf Solutions or did You develop them in the house we develop Our simulation house we believe that It's so core to our uh to our mission That we need to own it ourselves so we Have a phenomenal simulation team that That allows us to test every aspect of The software and simulation very Thoroughly across a wide wide range of Of situations that that the vehicle May Encounter you're nodding over there yeah I mean I I completely agree with that Approach obviously Um so all of our simulation has been Developed in-house Um and it's hard for in vehicle autonomy You you want to be making progress early You want to be getting Vehicles on the road that demonstrate The The credibility of the program but It takes a while to to build a Simulation program and so Aurora from The beginning has has been building this Simulation program and and so you've Really seen over the last couple years As our simulated miles have have started To increase exponentially to the point That we are very very uh kind of Prescriptive about where we use on-road Automated driving to uh to be kind of Feeding back and and building confidence In the Fidelity of our simulated system A lot of a lot of people out in the Audience there are probably thinking the

Simulation is is incredibly expensive up Front are there any solutions that they Could use to bring similar things In-house to their companies well of Course it does exist like companies that Provide simulation solutions to other Companies that they can readily use we Have deliberately made the choice not to Go that route but then to build Ourselves again for the for the because It's so important and so critical in Core to our mission that we to be able To we need to own that technology sure And this is very important we have built That from the very beginning okay let's Focus very heavily and invest very Heavily on building a simulation system Um so and we have not first gone to the Public road and then build a simulation System but the other way around and if We go on the public road we will do so Only after we have confidence from our Simulator right and the private test Tracks that we are really ready to do so And that we can safely deploy it of Course at first with uh with uh with a Safety driver but that we can actually Go on the public road and not cause any Risk to the public no we were talking Um with with my staff about this and Somebody brought up an interesting point That young drivers have to do a driver's Test and that's a standardized test Should there be a similar test for

Autonomous vehicles I believe so uh that Would be great yeah I mean I I think we Collectively both at a regulatory level And as a societal level we need to have Some agreements around what uh what Constitutes a safe system so I always pause because I worry a little Bit about like one performance does not Make a platform like you know I have a Daughter she's 15 she's going to get her Her learning permit quite soon and she Will at some point in the next couple Years be able to pass a driver's test And that will just be the beginning Right she will not be a safe driver Having passed that driver's test and so I you know when you look at the at Various Industries uh that build safety Critical systems you have to think about Holistically how those systems are built And and so I think there is you know the The foundations of what will make uh a a Autonomous vehicle without a safety Driver both safe and performant will one Of those things will be uh like Practical demonstrations uh but I think The process and production of the Software Um and and the way the hardware is Developed will also have to come under Scrutiny so I think there will be quite A few challenges in that like of course First of course there needs to be a Definition of what what constant safe

Like we of course Define our own uh Definition but Aurora does it too and There might there will be differences There so somehow there must be some kind Of understanding across the industry About what that is and then the second Challenge is okay how do you then test Uh like if there's some kind of Standardized test for whether you Actually uh who should build that test Exactly that is a big that is a big Question of course how uh how that I Don't know how that is going to play out Uh but uh I very much believe in the Philosophy of that approach so we are Doing that in-house and based on our own Uh definitions of safety and and what it Constitutes to pass that test uh but it Would be great I think in in general to Have regulatory like yeah guidance as to What what what a safe vehicle uh is and What what acceptable risk is what do you Think the current regulations in place Are robust enough to handle the Onslaught of self-driving vehicles now There is virtually nothing at the moment Right so Um like there needs to be some kind of Definition as to what does it mean for a System to be safe like in our particular Case driving on the highway like uh what Is the probability that we are like the Occurrence rate of let's say the the Vehicle departs the lane or something

Like that that must be a very very low Number obviously but that number needs To be defined somewhere I believe Um and we have we can make our own Number of that but if there is some kind Of this is just one example but if there Is regulatory agreement on what that Number is and then we can test towards The number that would be helpful right Sure we all held to a common standard uh Across the industry yeah I think it will Be an active partnership uh both amongst Industry within industry and and uh and With regulatory bodies yeah but it's not Going to be easy because of course There's a lot of intellectual property Associated with with what constitutes a Safe system like if somebody knows what A safe system is then that is extremely Valuable we have a I believe of course That we have a good idea of what that is But uh we may right and other companies May not want to share that with each Other and so there is going to be some Challenge how this is going to play out It's going to be interesting uh how to How this will how this will evolve both Of your companies have been in testing For a while let's talk about deployment When are these going to hit the roads When's the first long-haul Trucking That's the the million dollar question Of course when well when when we are Ready to do so right and we need to

Build a safe system and uh and that Meets all of the requirements that we Have and our approach generally is like We have a single fault tolerance system Across hardware and software that needs To be built and then we have Requirements for all the sub-level Components that need to be validated and Then we need to validate on the system Level that we can safely deploy the Vehicle on the road and if that is all The case then we will deploy the Difference take me through the the how Ike works or the solution you're going To provide Uh yeah so so we have requirements like Across the uh the the components of the Of the system uh like what they need to Uh need to be able to do and if we can Validate that they actually do that then We are going to Deploy that on on shirt and we uh have The data in-house but also external data About like the the frequency of of Events that occur on the highway there's Actually a good data set for that so That we can have an idea of like what Are we likely to occur and based on that Sample that distribution we test our System across variations of those Situations based on the statistics that That are there to see does our system Meet the requirements that we set for Ourselves very high bar and then we can

Deploy around the world what data do you Have that shows that that Americans are Ready for trucks to cross Texas without Any drivers Uh that's a good question Um thanks yeah Um I don't know if if there is data I Think this is a question that is open to A large degree we have to see where this Goes obviously if we deploy self-driving Vehicles whether that is in the city or Whether it's on the highway those Systems need to be safer than a human Driver right at least and probably much Better than that for the public to Accept such systems so I think this is Going to be a process uh in society Where the acceptance of these vehicles Will increase uh and and at some point It becomes the norm but how that exactly Is going to shape out is is not exactly Clear yet right it is still a very new Technology so I think all of our Companies are playing an active role in Trying to shape that that narrative and Then and and trying to educate the Public onto what it means but yeah this Can go in multiple directions and it's I Think our responsibility to make sure it Goes in the right direction correct and That people will be accepting of this Technology and the key is really safety Like this needs to be safer than human Drivers what about at Aurora so Aurora's

Focus the the focus of the engineering Teams is around the this product that we Would describe as the Aurora driver and So it's kind of all fun and games in Autonomous vehicles until you actually Build a system that can allow for the Removal of the Safety Drivers Um as we are developing this product we We also we're building it to be applied To multiple different businesses and so We're the The critical focus is around building Partnerships like we want to build Something that can be fielded by other Organizations we want to empower uh and Augment other other companies and Industries and so uh in in logistics and People Movement we want to build the Aurora driver and that's the Focus right Now is building both the driver and Defining the platform that that driver Can be applied to and and again like What um what redundant components are Are required for this system to be Controlled safely so that we can apply The Aurora driver both to uh to the Movement of goods and the movement of People so you know we're we we're we're Integrating this we've integrated this With a passenger sedan we're integrating With a passenger minivan where we've Integrated with uh with with trucks and With panel lands um and uh and so we're Constantly building this system that can

Be kind of uh adaptable to to lots of Different vehicles and lots of different Uh businesses and so and then Like we're not necessarily trying to Field this product as as a business we Own we are trying to partner with others Who can who can then use this to augment Their business is the current technology In the market good enough to make that Goal possible Um Well uh I mean Prior to uh Aurora's acquisition of a Lighter company in Bozeman uh I I I Would have said it it needs more time Um specific especially at highway speeds With uh with uh High control latencies Like you know large braking distances uh Large stopping distances for for large Vehicles the the range of a multimodal System that you want to apply for for Lighter camera radar Um I would have said it needs more time and So since we have uh have have brought a Fmcw lidar in in-house it really has Been a game changer for us in what what We believe we can we can uh how we can Operate with with with excellent Perception at range Um and uh and demonstrate uh validated Safety case around that Tesla says or People say that Tesla is six years ahead And you're smiling

Tesla says that they're six years ahead In technology what do you guys see well Elon Musk said that it would launch last Year I believe so yeah I don't think so but right yeah Well it's not here yet that's right are They six years ahead of Aurora Uh I don't I honestly don't know that There's any way to judge uh the state of Any any given program there's kind of This there there's this kind of zero to One moment right that we're all trying To achieve in terms of when when a System when you can actually demonstrate Not just the system can that is can Perform without a safety driver but that That over the over the the Long Haul uh Is is trustable Um And You know it's hard for me like I I Worked with the Tesla team I think They're they're great people right and We're all kind of uh we're all coming at The problem uh from different factors For instance they don't use lidar right No is is that the correct way to do Self-driving cars uh so I'm trying to Get you to talk about Tesla I would Never do uh no I mean I seriously would Never like uh I I have a tremendous Amount of respect for them Um what I would say is

The the most interesting challenge that I see right now is around motion Planning is around once you know what's In the world how do you decide uh how to Do something reasonable and safe in that World and uh you know this is this you Know right now we have a lot of Roboticists here and uh and getting a Sufficient ground truth to to make Motion playing decisions is is is Usually like the the big like the first Hurdle Um and so lidar allows us uh very very Quickly to to focus more on the motion Planning problem we get to we get to the Starting Gate of the motion planning Problem Um and really allow that team to to Explore the space very deeply at the Same time in parallel that we focus on The the long-term and long-range Performance of a multi-modal system Um so I you know I think you can absolutely Solve this problem with with cameras Only but uh but it's a matter of how you Want to how you want to break down the Problem and what time frames you want to Be tackling what parts of the challenge Yeah I think what's interesting about Their approach is that they've really Basically taken an L2 system and kind of Gradually like improving that according To their approach to to get maybe to L4

At some point and my view is that in the Middle of that of that transition you Cross some kind of uncanny valley where It's not clear like who is in control And that that creates in my view a very Unsafe system where people start to Trust the system even though your system Cannot be trusted and then you have These accidents that actually already Have occurred Um I think what what I find amazing is That Tesla has been able to somehow get Away got away with it and that that's What they're doing very well somehow Right like it doesn't stick to to their Brand as much because they have a few Accidents right right well I guess the Question is how safe is to say right Well I think you need to jump to The L4 Like at once like you'd not go through This Valley where it is not where where The system feels uh safe and it's safe Like 99.9 of the time but in the case It's not safe it suddenly is very unsafe Right to be really be safe all the time And where you can really rely on the System and not rely on the operator Anymore and that in that Gray Zone in The middle uh that that is potentially Very unsafe I think but I mean they're Doing it so uh and they're doing pretty Well with it it seems so well only a few People have died Yeah okay all right let's get back to

Technology I asked Jenna this earlier but Technology what would change for Ike if You had better technology or what's Missing to for deployment you know I'm Intuited to this point like about lidar Like when we started like the the range Of the lidar was uh was like you had Fellow dying and that could not see far Enough for our application on the Highway but like there were at the same Time it was an enormous uh yeah growth Of number of companies that do lidar Also at long range and so the timing for Us was perfect like the total technology Is now maturing and that is what we will Be relying on like long range lidar and Of course that's the better that that Becomes the better we will perform right And and but like in terms of timing it's It's great because I hope by the time we Are ready then they will be ready too And the conversion is nicely what about 5G Is that exact change the equation for Your solution potentially because 5G the Promise is that the that the the latency Of the connection is low and that would Allow more so than today uh Uh like remote control perhaps or very High fidelity uh remote assistance so we We are like keeping uh keeping a close Tab on what what is happening with 5G And uh and if that indeed fulfills the

Promise then we may exploit that Technology for this particular purpose Uh for our before our products we only Have a couple minutes left here and Talent recruiting is a big thing right Now in your industry How does small startups like yours Recruit against Tesla or Argo or General Motors what do you bring to the table What do you tell these people why why Should they come work for you well we Are still really small I believe like Compared to uh well they're a little Bigger but you're very small yeah so That's the excitement of it right like Everybody who works for us has an Enormous amount of ownership into the Problem we have a visibility into the Into the challenging problems uh a lot Of visibility in into how a company is Run and people are close and uh and and The the communication links are short so It's not overhead in terms of Bureaucracy Etc so it is an enormous amount of fun To work at a startup and you learn an Enormous amount compared to to a big Company I've worked for big companies And I would recommend to anybody even if Not for Ike go work for startup that's Where you learn the most then right yeah Really where you see the most and and Work with senior people as well as uh Your peers and it is just an awesome

Experience are you hiring right now of Course what are you looking for across The board uh Engineers but also policy And products right Etc What about Aurora we have a we have a Stand upstairs so if you're interested Then please So uh in any job search right I think uh You know I I focused uh strongly on our Engineering recruiting but obviously Aurora's hiring across many different Disciplines but especially For Engineers It's it's usually this exploration of Trying to find an alignment between Mission uh and and peers and uh and then The overall value proposition right so Most people come to Aurora kind of Self-select for the mission uh and that And that's and that makes for a very Exciting place to work when when people Are mission aligned it cuts through a Lot of the noise and and debate about Engineering Solutions and it kind of Breeds a a broader trust within the Group Um and then what people find when they Come to Aurora is also that like we have A very uh tight-knit and and as small as We can get away with Um a high Talent team right so we tend To hire very very experienced people Very senior people uh to try to allow Our teams to stay as as small as we can

Get away with while tackling enormous Problems Um and so that allows for a lot of of a Lot of interaction between between Really you know some of the top Minds in These fields and um and so it's a very Attractive place for people to come at Various levels of of career uh to be a Part of this and be exposed to this and And like we're we tend to be a fairly uh Humble and somewhat introverted uh Organization of Engineers um but Everyone's super generous super generous With their knowledge and their time with Each other and I think that really um That's frequently where we win in in Competition uh with with other people in The market is the the creative and uh And learning environment that we can Provide Um and then obviously you know the Overall value proposition is is Important to everyone so kind of where The kind of what people think they can Accomplish what they think they can Learn and what uh what the overall value To them will be for their time at Aurora We're out of time but I'm going to steal One more minute Um I had questions about SpaceX because You just came to Aurora from SpaceX After spending about 10 years at Google Yeah so I guess the last question is Who's a better boss Chris ersman or Elon

Musk No they've all taught me uh Like a tremendous thing a number of Things that I never knew right like I've Grown I've grown deeply under just about Every boss I've ever had I've worked for Chris Armstrong he's great there we go We'll take it with that well thank you Very much guys I really do appreciate it Give them a round of applause Thank you