Formula 1 has followed up the launch of its OTT TV service in May 2018 for the Spanish Grand Prix by starting to migrate most of its video and data infrastructure from in-house data centers to Amazon Web Services (AWS).
The two moves are intimately connected in that the two OTT services, F1 TV and premium option F1 TV Pro, allow subscribers to personalize how they watch a Grand Prix race, while AWS is providing machine learning and other capabilities that will help deliver that promise. Subscribers can select camera views so that they can home in on particular drivers and select data they want to look at, which will now be delivered from the AWS cloud.
Formula 1 was swayed partly by the scale economies of AWS but also by the promise that the Amazon SageMaker managed machine learning service will generate some new insights such as whether a driver has just performed an exceptionally good lap for his ability and the prevailing conditions which can take account of factors like tire wear.
One interesting aspect of this is how it seems to overlap with what individual teams such as Mercedes, Red Bull and Ferrari, are doing to increase their competitiveness. In the past it has been all about engineering expertise to squeeze an extra unit of engine performance within the ever-evolving rules imposed by Formula 1. Now they are adding data analytics to their armory, both to extract small advantages during races by optimizing pit stops and tire types, as well as feeding back into engine design.
But this same data, combining historical data collected over 65 years with proliferating real time data generated by cars during the race, is also now being harnessed for viewers via AWS. Formula 1 is now training deep learning models with SageMaker on that historical data now stored in Amazon’s DynamoDB and Glacier cloud storage service.
This provides both the performance statistics relating to past races that can be extracted and the bedrock for applying machine learning in combination with real time data in split second decision making such as when to make a pit stop for maximum advantage, or whether to change the tire plan for a race.
Real time data currently collected includes telemetry data from onboard sensors measuring speeds, vibration frequencies and sizes, temperatures and tire pressures. There are also measurements of driver input to steering, acceleration, braking and on top of that the verbal communications which has potential for natural language processing and analysis. There is also track data captured at intervals such as 100 meters with lap times, maximum speeds per lap, pit stop times, track temperatures and wind speeds.
Some of this data such as maximum lap speed is of more interest to viewers, while other data such as track temperatures may be more relevant to the teams themselves. We can only presume that the insights available to the teams will be greater than those to viewers or else there would be no competitive advantage. But one issue here is that huge computational power is needed to drive the deep machine learning algorithms that would produce advantageous real time insights valuable for decision making within the pits or by drivers themselves.
Formula 1 has resorted to AWS largely to obtain such power, so it will be in the data that teams will hope to get their advantage. It may be that Formula 1 will ration the insights which may only affect the more nerdish race addicts who are just as interested in obscure parameters as the teams.
As well as the analytics Amazon will also be taking over F1’s video asset workflows through AWS Elemental. This has around 8 components that span the workflow from ingest to distribution. AWS Lambda starts the process by triggering various so-called AWS Step Functions for ingest, processing and publishing workflows. The first Step Function ingests a source video, with or without a metadata file, validates these and then generates its own metadata used for the rest of the workflow. Then a second Step Function generates an encoding profile based on the metadata and submits encoding jobs to AWS Elemental MediaConvert, which applies two-pass encoding to generate multiple versions of the original file. After encoding is complete, a third Step Function then validates the output.
Both source and destination media files are then stored in Amazon Simple Storage Service (Amazon S3) and file metadata is stored in Amazon DynamoDB. Amazon CloudFront then handles global distribution of the on-demand video to end users. Meanwhile Amazon CloudWatch monitors log files and sends notifications for both publishing and handling errors.