You might think that auto racing would not be a good application for quantum computing because the teams consist of grease monkeys who may know auto mechanics but wouldn’t know how to leverage advanced computing. But you would be wrong.
Auto racing is a big business where there can be a very thin line between success and failure. To give you an idea of how small things can make a big difference you can look at the results of the 2015 Indianapolis 500. In that race, the difference in finishing time between first place finisher Juan Pablo Montoya and second place finisher Will Power was 104.6 milliseconds. And those 104.6 millisecond made the difference between winning a first-place prize of $2.44 million or not.
It turns out that an auto race generates a lot of data, about 1 Terabyte per car in a typical race, that if analyzed and used wisely can help give a racing team a critical edge. To that end, Zapata Computing and Andretti Motorsports formed a partnership earlier this year to work together on race analytics and see how they could use Zapata’s advanced analytics, quantum techniques, and Orquestra hybrid classical/quantum data and workflow manager to win more races.
Although this work between the two companies has just started, a big event for both companies will occur this weekend with the 2022 Indianapolis 500 race. We talked with Christopher Savoie, CEO of Zapata Computing, and he described three of the first use cases where they believe advanced analytics, machine learning, and quantum computing can potentially make a difference.
Tire Degradation Analysis
When you have a car going at over 200 MPH, the tires wear out very quickly. In a typical Indianapolis 500 race, the tires can be changed 5 or more times and require time wasting pit stops to accomplish. What’s more the tires have different characteristics when they are just put on and when they have been used a while. So, the racing manager has a lot of strategic variable to juggle. When should the car be called in for a pit stop to change the tires, which set of tires should they put on the car, and how many tire changes should they have, and what is the current weather and track conditions? For a data analyst, this is a large optimization problem and will be one of the first areas that Zapata will work on with Andretti to create a ML model that can help guide these decisions using data collected in previous race sessions as well as data collected in real time during the race.
Fuel Savings Opportunities
Cars need to be refueled during the race. In addition, the driver has some control over the fuel consumption by the way he drives. If a racing team can find a way to minimize the number of refuelings and avoid a pit stop, it can save a lot of time. What’s more you don’t want to cross the finish line with a full tank because they would be a waste. In the 2016 race, driver Alexander Rossi took a gamble and decided not to go for a final pit stop to refuel with 33 laps to go. It turns out he ran out of gas at the very end and coasted across the finish line. But he won the race because the second-place guy did decide to refuel and the extra pit stop time cost him the race. So, finding ways to improve fuel consumption and determine the best timing for refueling also turns out to be an optimization problem that may an opportunity to use machine learning and advanced analytics to find the best solution and improve race performance.
Yellow Flag Predictive Modelling
A yellow flag during the race occurs when an accident occurs or there is debris on the track. Drivers are required to reduce their speed and passing another car is prohibited. One of the impacts of this, is that the relative lead of one car over another is reduced. But it may also be a good time to go in for a pit stop since the cars aren’t going at full speed while the flag is on. If a racing team had a crystal ball and could predict when a yellow flag would occur, it could help them determine their best pit stop strategy. This may seem a little far-fetched but the Zapata/Andretti team will attempt to create a model for this that will be based upon conditions on the track, the status of the various cars in the cars, which particular drivers are in those cars, and other factors collected during the race. It will be interesting to us to see if they can actually create a useful model for when yellow flags may occur from this data.
From an operations standpoint, working in this environment can present some unique challenges. But it also provides learning opportunities for the Zapata team as they face real world challenges and find ways to solve them that can be used for future product enhancements and customer engagements in other areas. One of the first things to understand is the racing environment requires real time decisions and you do not want to use a quantum computer somewhere in the cloud on race day. The latencies will be too slow and you don’t want to have to struggle with flaky Wi-Fi connections. So, Zapata and Andretti have set up an on-site Race Analytics Command Center as shown in the picture below.
Zapata and Andretti aren’t going to install a quantum computer in this trailer, but it will have a large amount of classical computing capability to help the team make real time decisions on race day. Machine learning applications are typically divided into a training session that develops the optimum coefficients for a model and an execution portion that just runs the model and provides an output based upon the previously setup coefficients. The training portion is the most computationally intensive portion of an ML model, they do not have to run in real time and is a good opportunity for leveraging quantum computing. Executing a model once it is created is not so computationally intensive and can be done on a classical processor. The team can feed in data from previous races and trial runs, create an ML model over many days or weeks, but then execute the ML model in real time on classical computers sitting in this trailer.
The collaboration between Zapata and Andretti goes much beyond leveraging quantum computing. The overall program will involve working with multiple data bases that could be resident with cloud providers, edge computing data coming in from various sensors, and managing workflows that are both classical and quantum in nature. Zapata will be using their Orquestra product to help manage all this.
This will be a long-term collaboration. Because the available quantum computers are not yet powerful enough to provide an advantage, the first implementations of this work will use quantum-inspired algorithms. However, the intent is that as the quantum processors become more powerful, these algorithms will eventually be moved for full quantum computers and allow the companies to create larger, more complex, and more accurate models to further their advantage. Andretti participates in many different types of auto racing and has many different teams. So, the two companies will have a lot of opportunities to try out and develop this capability. We also expect the companies will find additional use cases for leveraging advanced computing capabilities as they work together.
For additional information about this collaboration, a news release posted on the Zapata web site can be accessed here.
May 26, 2022
