Overview

pGenpact is a global professional services firm that makes business transformation real. We drive digital-led innovation and digitally-enabled intelligent operations for our clients, guided by our experience running thousands of processes primarily for Global Fortune 500 companies./p pThrough our work with Envision Virgin Racing, we combine the race team's engineering expertise and our skills with digital technologies. With faster access to accurate predictive insights, we help the team make lightning-fast decisions for superior performance. Our work has revealed important lessons in digital transformation for race teams and Fortune 500 companies alike./p

Supernova Award Category

AI and Augmented Humanity

The Problem

pLike many organizations Envision Virgin Racing was looking to optimize their limited resources and come out ahead of competitors.The biggest challenge in Formula E racing - where teams compete given finite battery power in 45 minutes + 1 lap  is to finish first without running out of energy. In contrast to many enterprises, which have time to plan and explore the different strategies to achieve their objectives, racing requires fast, intense and on point decision-making while competing with 23 other drivers at 170 mph. On top of keeping an eye on competitors' every move, the driver must also watch out for track conditions, environmental factors and avoid being fined for breaking certain regulations./p pWith so many factors to consider, it’s essential that teams get the fundamentals right, In Formula E, as 45 minutes whittle down the clock in a race, the drivers who come out on top are those who have an edge in estimating the number of laps remaining. Underestimate, and the driver ends up overspending his energy; overestimate, and he will be too conservative and not being competitive. The challenge addressed here is to estimate with precision, taking on board a huge number of variables and historical data./p

The Solution

pOur solution needed to augment the technical engineers’ strategy in real time, and produce accurate predictions through machine learning, fed with timing data. Usually, evolutionary algorithms are slow and require a top-down approach to yield sensible outputs, deterring end users from leveraging them real-time. This use case challenges that typical perception, and in extreme conditions out-delivers traditional methods on speed and accuracy./p pOurstrong Lap Estimate Optimizer (LEO)/strong is an ensemble of evolutionary algorithms that explore the best fit to the racing dynamics at hand, given the time remaining, leading driver’s speed, and (safety) conditions, among several factors. Ranging from simple iterative models to symbolic genetic algorithms, the models compete towards reaching the right estimate sooner, across different scenarios, and continually update a consensus output as the race progresses./p

The results

pIn a racing context the value delivered is clear: accurately guess the number of laps remaining, and the likelihood of winning the race is increased because the team can plan their actions better. LEO has proved that machine learning can support this challenge especially well in changing track conditions, such as a drying track or a track accumulating water from any rain, In such circumstances LEO has a clear edge against the team’s classic methods, And in Santiago, a very hot race, strongemLEO settled on the right number 15 laps earlier/em/strong./p pAdditionally, LEO has helped the team manage their energy more effectively. Since we started working together to develop strongLEO, Envision Virgin Racing has not run out of energy short of the finish line/strong, unlike its competitors. In business, employing the same optimization enables teams to achieve better collaborative efficiency, which in turn allows for a more agile way to innovate, respond to changes, and grow the business while protecting the bottom line (i.e., the definition of a winning charter)./p

Metrics

pIn racing, the key metric is binary: win or lose. And it is a very public metric to be judged against every two or three weeks./p pLEO, in conjunction with Envision Virgin Racing’s in-house strategy tools, enable Envision Virgin Racing race engineers and strategists to be better informed to make key decisions in a race (e.g. energy and tyre management)./p pOn two specific occasions in the 2019 racing season, the insights that LEO provided, coupled with in-house tools, led to victories on the track./p

The Technology

pstrongPython-/strongThe Optimizer’s algorithms are coded using libraries like sklearn, pandas, numpy. strongMongoDB-/strongTostrong /strongstore data flowing into the machine learning algorithms and to direct the predictions from LEO and neural networks into separate collections for the BI platform. strongRShiney-/strongTo automate race stint analysis and to automatically generate multiple analysis. strongNeural Networks-/strongLong Short Term Memory based neural networks to predict lap wise timings, Montecarlo simulations algorithms to generate data./p

Disruptive Factor

pThere were a few challenges that were faced both in the design and the implementation of the solution./p pOn the design side, the first challenge we faced was around the integration of data from multiple data sources (Track, weather etc.) into a consolidated data model. It required data engineering expertise to envision and design the data model up-front and establish the pipelines to ensure assimilation into a data lake. The second challenge was around the latency faced into such integration since it was key that it happens real-time. This was addressed with the establishment of a Kafka server that provided a unified, high-throughput, low-latency platform for handling real-time data feeds, and, optimizing the codes to that end./p pTwo key challenges in the implementation process were (i) the requirement to rapidly learn how data impacts the business of racing and (ii) the incredibly short lead time to build a working prototype. Such are the tiny margins of performance that define success in racing, the process of developing competitive advantages is relentless. The impact of the project was significant in that LEO gave the team additional insight into data and provided confidence to make quick decisions that could be the difference between winning and losing a race. Embracing this technology in Formula E was ground-breaking and has helped transform Envision Virgin Racing into the world’s first instinctive racing team./p

Shining Moment

pLEO gives superior insight for every race, so it can make the right decisions when a safety car pauses a race, the weather turns, or when a competitor attacks. LEO's insights are particularly helpful when faced with changing track conditions. In Paris where the team’s driver, Robin Frijns succeeded to take his first victory. LEO has consistently predicted the actual number of laps ahead of the existing technology, notably in Santiago, it settled on the right number 15 laps earlier./p