Curriculum - How do autonomous vehicles work?

how do autonomous vehicles work [Music] hey everyone i'm super excited to be here today with adam and we're going to be doing a mock tpm interview um in this particular interview we're going to be asking a question about autonomous vehicles but before we get into it we'd love to have you adam tell us a little bit more about yourself absolutely my background is in mechanical engineering i've been a technical foreground manager for about a decade now worked across several different industries including the government consulting and also now tech and most recently uh interested in autonomous

vehicles and happy to be here and chat with you about it today awesome well we're super excited to have you at exponent we believe in bringing the interviewing community together to make a more empathetic and guiding space and so it's great to have an expert like yourself on the show today so without further ado let's jump into it in today's question what i'd like to ask you is how do autonomous vehicles work yeah the first step of that question really requires knowing the industry standards what we call the sae levels of autonomy so there's actually

six different levels starting with zero up to five and i'll talk through that briefly the level zero vehicle is what you see on the roads today people driving to and from work going about their commutes and habits and then sort of level one two and three are hand in hand and what we see as you go higher up the stack is more and more the system is driving the car rather than the person so example of a level two or level three on the advanced side would be a tesla autopilot or a cadillac or general

motors super cruise where now instead of having to be fully attentive on the car at all times we're getting more and more that the car is driving itself uh how i'd like to answer your question today is actually the level four of autonomy and in that case the biggest difference is the car is driving itself and also there's backup systems uh in case of safety so the final ultimate level and this is sort of a pipe dream for many different companies in the industry today is level five autonomy where a car is going down the

road by itself there's passengers or cargo or or whatnot and there's a secondary safety system but also we're not constrained based on weather and we're not constrained based on geography but again i'm gonna focus on a level four and to do that you know a lot of companies do things differently but i'd like to split out and break this down into talking to you first about the vehicle then about the sensors then about the software sac so the vehicles themselves today you may see test uh vehicles out on the road and largely they're just like

the cars that you buy and use today with the biggest exception that they have additional space for the autonomous vehicle hardware you know a lot of these cars have safety drivers and so the systems may be driving the car and there's someone there as a backup but you know the vehicles are pretty similar to what we can buy a lot of companies choose to do electric vehicles and that's a big push right now because when we move to autonomous vehicles and and push out we have the opportunity to switch over from combustion engines to electric

which can have some benefits both for the environment and also how these will operate so the second tier that i'll break into is the sensor and specialized hardware on the vehicles um the biggest thing that pops out when you see one of these rolling down the road is the lidar so the spinning laser light and finding so autonomous vehicles have many different sensors because they're trying to replace the driver lidar that's the one that has range finding and you can see how far away objects are from you a big challenge with lidar though is having

to do with weather and sort of being able to see in degraded environments of fog or rain are difficult for lidar cameras actually i would say most people would agree that cameras is the most important sensor today and the reason why is that all that imagery comes back and is fed into machine learning mod models so that we can classify a nurse and when we see a person we see a dog when we see a car and you sort of want to understand uh what you see but the card and system know so that we

can then determine uh what to do next uh the third really popular sensor is radar now radar is great for cutting through weather you can also tell how fast objects are going but to be honest it doesn't have as much precision and granularity as cameras and light are so you can see something big on the freeway rolling down the road but you don't actually know if it's a semi versus a car uh so we have all these different sensors that are specialized and i'm just trying to give the car and the system the best chance

to understand what's going out um in the environment the other popular sensor that you know is actually used on cars today is ultrasonics you'll see those in bumpers this really helps with very short sorry very slow speeds is where ultrasonics comes into play this will allow a vehicle to park and also see any sort of shorter um obstacles or people around and then uh you know there's a lot of talk now about using thermal cameras uh to try and see better in the night uh finally one sensor that people actually are less familiar with is

that a lot of autonomous vehicles today are starting to add microphones now you would wonder why do you need a microphone and that's so that when a police car or an ambulance you can actually uh hear and identify the siren or you can listen to instructions from someone in the street and yeah that's why microphones are used today the third piece is the software stack so now that we know about the vehicle and you know about the sensors what are we actually doing with this information and i sort of think about it as breaking down

into the different components of the software stack and this is where the real magic is to make an autonomous vehicle work first you receive that sensor data and you have to process it you have to do something with it um you have to turn those still images in order to understand who or what is around you then that is fed into perception once we get a profile of what the object is if it's a person or a dog we those objects then have different behaviors so a person may be waiting at the side of a

corner and know to stop for a car a dog however might actually just run straight across the street um so this sort of the perception stack has a way of figuring out what is going to happen next uh one piece i haven't really mentioned yet is mapping um yeah i think people think that uh autonomous vehicles today can sort of go out and and be ready to roll into any city but that's not the case we actually have to send an advanced mapping car so that the autonomous vehicle can then check against this map and

know where it is at any given point and this goes back to lidar and the real benefit there after a mapping car has gone down the street identified certain signposts and banks or certain uh sort of still locations when the car goes through it will then again check these and see these and know where it is at any time i mean yes there's gps and inertia sensors on the vehicle sometimes for the same localization purpose but mapping is still required in a lot of these test vehicles going back over to the perception side once you

know what is out there and where it's going to move next you have your own motion planning because you always have to be as the vehicle steering around the different things on the road so if a person is standing on the right hand side and you want to make a left turn you know you can safely execute it and the vehicle will know to do that left turn safely either if you have to wait for the person or not and sort of going all the way now to um you know that's the nitty-gritty of the

software but we have to get the car to know what's going on once you have your motion planning you actually are then connected to the cars controller and a system called can bus that's how we give instructions to the car about where to move and actually apply the gas and it's not that there's a robot sitting in the front seat with a steering wheel and and operating these pedals but you know we're talking machine to machine to get the car to move so as simple as that was that is how many autonomous vehicles work today

these three pieces work in concert to hopefully one day have us a driver who's fully attentive and can get a safely to work wow adam that was a great explanation of autonomous vehicles and thanks for going into all that um i do have a couple follow-up questions that are more related to technical program management as it relates to autonomous vehicles so i'll go ahead and dive into those a little bit um i i want you to imagine now that we've had an excellent explanation of autonomous vehicles let's say that you're a tpm at an autonomous

vehicle company or you're working with autonomous vehicles what are the main trade-offs that you would have to consider that you would feel important to consider as a technical program manager yeah great question as a technical program manager i think the most important thing to do is managing the cost schedule and sort of quality and performance of your project an example i'd like to use is if you're a hardware program manager on autonomous vehicle and you own the compute module or any sensor module in that case a lot of the times you're working directly with suppliers

and these suppliers want to have a good idea of the quantities that you're working with what they have to build for you but a very specific trade-off that you have to do is do you make a system more robust or do you make it just good enough for the application and i'll tell you in the companies that i've worked with uh it's no nonsense the number one is safety so we will often see and i would classify safety probably the most into quality and performance that schedule can be delayed and costs can go up as

long as we're maintaining safety and improving the lives of the riders totally yeah it's such an important aspect of this type of product that can be different from other types of products safety is not really a consideration in other product areas um i want to go on to the next question that i had just another quick follow-up question around again imagining that you're that tpm for this program what are some communication formats that you might consider when running this program or how would you communicate across the team yeah for that i'd have to look at

the tpm toolkit and you know there's a lot of different tools out there today first and foremost the thing that pops into my mind is sort of regular email updates that you can send company-wide or to your targeted stakeholders about your program you want to be letting folks know especially about your schedule and risk that you've identified and sort of if you run at any obstacles and need to pull in and help other ways is obviously in tickets themselves so popular tools today asana jira you know you're always communicating and using these different ticket reference

numbers because they have the whole history of whatever the issue is so whether you're sending it in a direct message to somebody or emailing out tickets are a great way to only track what your work is and what the priority is and also communicate that to others and then of course there's meetings and we've been doing a lot of zoom and online meetings today you can set different meeting cadences for that you know in my past projects i've presented to the ceo and cto on a weekly basis about what the status of this critical part

is and you have to tailor your communication style there a little bit executives want to know um just what the sort of high level points are and how it's going to impact the program as a whole and then you have all of your regular syncs with the individual contributors on your team so now this is when a tpm has to roll up their sleeves go in with the engineers and you know really just let folks know what what is blocking uh who and sort of what the priority is any given point so i think within

all these different tools in any great program will use maybe all of them or some of them based on what the complexity is awesome well adam this has been great i want to move on to talk a little bit about feedback and you can kind of take off the interviewee hat in just a moment but is there anything else you'd like to add to this particular interview answer uh no i think that that really is what it boils down to um for this this type of questions i think it's really go back to the fundamentals

of the tpm function itself and make sure that you're covering all the bases because the most important thing that tpm can do for their program is one really clarify communications to be a constant you know positive force and reinforcement for the whole team and three ultimately uh you have to having someone be accountable for the success of the program these three things are what make tpm's great totally yeah um overall excellent answer adam this is super awesome to hear and i think will be super helpful to all the tpms watching um i do want you

to take off that interview we had and maybe we can reflect together about um you know i know you just gave a little bit of your self-reflection i i wrote some notes about what i thought you did really well and what i thought made this answer such an awesome answer um one i i thought you you used frameworks judiciously throughout this answer so you brought in like tpm toolkit or hey cost schedule quality and performance or factors i'd consider but you did that in a very uh in not a very overbearing way so sometimes i

often recommend interviewees to avoid using frameworks because they can often rely on them too much in a very natural cadence you're talking through things you're enumerating them like this is the first point second point third point um and you did a great job of um sort of like setting level setting like where we are and making assumptions about what i was looking for we're not looking for a level zero to level three vehicle we're looking at a level four or five and kind of discussing those and making those decisions yourself i thought that overall led

to a great answer i know we're in a little bit of a contrived setting but the only piece of feedback i had would just be checking in with me a little bit more as your interviewer it's always a conversation not just a one-way direction and sometimes it's easy to kind of get carried away i i personally thought it was so engaging and you did such a great job that i almost didn't really want to interrupt you but in an actual interview setting it's always helpful to check in to know how long the interviewer actually wants

you to go but overall i thought this was a really awesome answer you did a great job articulating your perspective on this and you clearly you're an expert both as a tpm and in the autonomous vehicle space but i did want to leave room if you had any other comments reactions maybe even tips to folks watching that you wanted to share at the end of this video well certainly i agree with everything you said including this uh critical piece of feedback um yeah it's it's a different world right now than uh when we've been interviewing

you know earlier this year doing zoom calls and interviews it's challenging sometimes to get that back and forth in but yeah you bring up a great point that it's good to check in sometimes an interviewer is asking a question and they want sort of a one-minute response sometimes they want something longer and you know you look for cues from the inner viewer based on how complex and how deep you should go um and just like you mentioned yeah it's always good to check in because you know i could have stopped at autonomous vehicles vehicle sensor

and software stack and then only then once an interview wanted to dive into one of those disciplines go deeper um yeah i agree with all that and i would say for tpm's uh like you were mentioning frameworks are very important because they're looking for you once you join the company uh to see that you can apply those things and throughout the interview process whether it's coordinating with a recruiter whether it's uh answering questions directly with people who are going to be your partners uh the next day um yeah no knowing that you can not only

know what these frameworks are recall them but actually use them fluidly is vitally important totally adam and just for folks watching i think a natural follow-up question is where do you learn about these frameworks or what are some of the frameworks that you think were most effective for you in interviewing uh honestly uh with a tpm uh the best way that i've learned is on the job so i think for me personally having this unique experience uh working again in government consulting and in tech now has really made it easier but you know you need

to be talking with other tpms i feel like that's what's missing today there's you see online that there's a lot of resources and groups and prep for product managers uh there's just really isn't as much out there for technical program managers uh we do have certifications like the pmp which i'll be honest i'm studying through now i haven't attained it before um but that teaches a lot of sort of by the book and these different structures but uh yeah the best way i've learned is working and also meeting with my peers and colleagues awesome and

we'll be adding uh some of the notes from adam's conversation today to the lessons in this course as well and just again thank you adam so much this is super super valuable for other tpms out there and i appreciate you helping us grow a community of tpms to help prep and ace the interview yeah absolutely thanks so much for your time today and to all of you out there good luck great good luck on your upcoming interview