Interview: Matt Harris, Head of IT at Mercedes-AMG Petronas

For the past 10 years, Matt Harris has been the Head of IT at Mercedes-AMG Petronas. This means Matt and his team play a pivotal role in helping the team to victory, in a sport that so heavily relies on IT, having the best team and the best hardware, software, services and networking means they have the edge over the competition.

I was fortunate enough to be able to put some questions to Matt and he was kind enough to spare some of his time over the Christmas break to reply. Enjoy the interview below and please leave any questions in the comments below.

When you head into the development of next year’s car, do you always trying to add additional telemetry points, or are you comfortable the set you have now provide the full picture to make accurate decisions from? 

Sensors on the car are weight, and every 100g of weight we add to the car can be up to 3 tenths of a second of lap time lost. So we’re always very conscious of adding a sensor to the car to help performance or reliability.

If it’s not giving us either of those, or if it’s not giving us enough of a performance improvement for the weight of the sensor, we won’t use it during the race.

We might use it during practice on a Friday or during our test sessions; sometimes you’ll see a car with huge aero rakes that add kilos of weight to its body, but they’re there as measuring devices for specific sets of information gathering and to test the efficiency of various components.

We do pick and choose when we use sensors, but we don’t necessarily add them to the vehicle – for the race, anyway – unless they’re specifically related to performance.

The sport is increasingly under pressure to maintain affordability. Are Mercedes realising any cost savings by analysing data through AI, rather than human time?

Mercedes’ senior management team sees AI as just statistical analysis on better computing and remain quite sceptical of how machine learning can help us.

This year we’ve had some of the first cases where machine learning has shown us something that the human didn’t actually program it to find, which has helped us improve processes. So while there aren’t necessarily any cost savings, there are certainly time savings.

It also means we can take longer to do other pieces of work more thoroughly, because we’ve shortened a piece of work that has to happen before we can start the design of next year’s car. Machine learning has reduced the duration of this piece of work has from 60 days to 30 days and is likely to be even shorter next year, so we have more thinking time/design time after a decision’s been made.

RoboRace is a new category that aims to highlight engineers over drivers. How do you feel about a goal like that, given so much of the work is done behind the scenes? 

For Mercedes, engineers are just as important as drivers. We take thoughts, comments and opinions from both sides, and we then use data to validate these feelings, whether from a driver or engineer. Neither is more or less important than the other.

In certain instances, we’ll prioritise the feedback or information from either the driver or the engineer – e.g. during a race when a driver needs to react to something, their feedback is critical for us to improve their experience moving forward. But engineers and drivers are equally as important as far as data is concerned. 

You’ve spoken about the need for ultra-fast data transfer between the vehicle and the team for analysis. Can you speak more about that challenge and how you overcome it, particularly in the heat of qualifying. 

Retrieving data from the car has always been our most important function from an IT perspective. We want to make sure the engineer can access the data as quickly as humanly possible. The data then needs to be analysed, decisions need to be made and the car must be turned around as quickly as possible.

Qualifying is the heat of the moment for us; we must be sure we’re making the correct decisions particularly during practice sessions P2 and P3, when we have a couple of minutes to turn the car around and re-fuel it before it heads out for its final session.

We tweak the car based on the data the driver’s fed back – including changes in track temperature, weather conditions, tyre pressure etc. If it takes 5 minutes to retrieve this data from the car, this is too long. All our sponsors and systems are working to ensure the supply of this information is as quick as possible, not only to engineers but also to multiple people globally, including those in our factories. 

Mercedes has been incredibly successful over recent years, what tools, technology or techniques do you employ to stay on top of the competition?  

We focus on continuous improvement in all areas. We’re very critical of processes and procedures. When we win, we don’t just accept winning as being the best we could have done – we still look at what we could have done better. And when we don’t win, we make sure we don’t make same mistakes twice – week on week, year on year or within a season. Our success is very much based on the actions of people, as much as it is on technology or processes.

What is your approach to developing your AI/ML skillset inside the team? 

18 months ago, we employed our first data scientist within the team. They were quickly swamped by too many different interests and requirements across the business of how to do things better – so we worked with our sponsors and partners to augment the team and embed the tech as a centre of competence within the company.

This team is now trying to standardise how we do things across the business and spot where we don’t manage or acquire data correctly or accurately. Now that we’re beginning to get this data in a better form, we’re starting to ask and answer better, more relevant questions – but we still have a long way to go.

You’ve mentioned before that AI is helping inform decisions, but not yet making them. Is that just a matter of time, or would see a competitive advantage to those who can make that leap first. What would you need to see to give you confidence that it’s time to hand that over. 

We currently use AI to help inform our team members or reinforce human decisions, or to give us the ability to make those decisions faster.

To get to a point where the machine actually makes the decision is much harder in the racing environment due to the number of variables. We don’t necessarily have enough data for the ‘computerised’ version to understand enough about what’s going on.

Machine learning and AI could start making decisions as part of the operational technologies within the business and automating self-healing of systems, which we’re already investigating within our CFD environment to make sure systems are available and performing sufficiently.

So away from the track, we’re at the point where we’re starting to trust these technologies – but on the track, the fidelity of data doesn’t allow us to trust the ‘machine’ answer sufficiently – in the same way we often second-guess our own self-made decisions. But having these technologies available as extra information sources can only help us.

Jason Cartwright
Jason Cartwright
Creator of techAU, Jason has spent the dozen+ years covering technology in Australia and around the world. Bringing a background in multimedia and passion for technology to the job, Cartwright delivers detailed product reviews, event coverage and industry news on a daily basis. Disclaimer: Tesla Shareholder from 20/01/2021

Leave a Reply


Must Read

Latest Reviews