ROB SMEDLEY
The former Jordan, Ferrari and Williams engineer explains his role in Formula 1’s partnership with Amazon Web Services, creating the ever-expanding range of performance and strategy graphics which accompany Formula 1 broadcasts
Explains his part in F1’s tie-up with Amazon Web Services
How useful is it to break down the complexity of Formula 1, telling the story to fans easily, and what feedback have you had from fans about those graphics?
We often have to ask ourselves when we’re making these graphics, “Is this just information that interests somebody like me and colleagues of mine on the F1 pitwall, or is this stuff the fans are going to want?” You’ve got to listen to fans… listen to what the customer wants and then work back from there.
Some stuff we were sceptical about, such as the corner-analysis one, showing the speed trace and how two drivers approach. In fact, the feedback on that just went off the scale. People absolutely loved it. And it had been a little bit the same with the braking one as well. Maybe you [as a viewer] don’t understand the physics of it, but you can properly start to appreciate what gladiators Formula 1 drivers are.
What are the processes you go through from getting the data and then deciding how it becomes a graphic?
We come up with a shortlist of between five and seven new graphics per year. My role is to bring that knowledge and know-how of engineering and data, and the technical and sporting aspect, of how F1 kind of knits altogether. We do a bit of a brain dump – me and other technical people in Formula 1 – and say wouldn’t it be great if, for example, we had this which we used to look at on the pitwall, or that’s a really key bit of information which is mega-important.
Then the TV production people get involved. They may have a different slant on it as we approach it as engineers. If I look at the graphics, they are nowhere near as complex as you would get on the pitwall. So you need that layman view of what the story is. Then we come together with AWS [Amazon Web Services] to reach the shortlist for the new graphics.
And then you get into the technical process. Essentially that’s my team, all ex-f1 team people like me, who have that sporting and technical knowledge, coming together with the data scientists from the AWS machine learning solutions labs. We build out all the mathematical and physical models, which have the machine learning algorithms. It’s the same process as in an F1 team – we’ll do loads of iterations of it where we correlate either back to real life or to the simulation environment.
It’s easy for people at home to be cynical about how accurate the graphics are – is there a credibility gap to be bridged there? How far apart is your data from what the teams are getting? The data set we have here is really rich. We [FOM] have got 25 to 35 [timing] loops, depending on the track length, and you can glean a lot of information through that. Teams are able to take a lot of competitor analytics from just three sectors. Twenty-five loops is an order of magnitude more information.
Like all data you’ve got to know how to use it. It’s not a case of “I’ve got more data therefore I’m better.” That’s where the partnership with
“I STARTED GETTING QUITE A FEW MESSAGES FROM MATES IN TEAMS SAYING, ‘HOW DID YOU DO THAT?’”
AWS comes in. You need that big data analytics and the machine learning element to bump you up to the next level.
So from a certain point of view, we’ve got more data. The car data is interesting, because you’ve got less [granular performance] data than the teams. We take a subset of the data off the car, and it’s essentially the subset you see on the TV graphics: lateral acceleration, longitudinal acceleration, etc. That’s driven us to build out some interesting car modelling techniques.
It was quite interesting, because when we first started this, there were a lot of comments: “How can they say that? Come on!” But then I started getting quite a few messages from mates in teams saying, “How did you do that? Because we’ve looked back on it, and it’s pretty accurate.
Are there great ideas you’ve had for graphics that haven’t worked when you tried implementing it? Or ones that aren’t possible to do because they’re too complicated?
Quite often, when we first do the “show and tell”, we have to come back with modifications because they’re overly complex. It’s been a real eye-opener for me, because as an engineer on the pitwall you’re always trying to absorb as much as you can – you want as many noughts and ones as you can possibly get. And it’s nice to see a different point of view on that – the editorial point of view, if you like, which is to say, “that’s never going to work, people aren’t going to understand it”.
So rather than trying to give 10 pieces of information that you’ve got to use, 20 years of experience to knit those together and come out with the right answer, we use machine learning techniques to try to get down to a much more elegant solution.