Features

Cracking the code to win the Cup

© BAR

The long-awaited revolution in Artificial Intelligence (AI) is finally here.

It’s impossible to look at business or technology news these days without seeing a mention of a new application for AI, or a new start-up company to take advantage of it.

From the chess-playing Deep Blue to driverless cars and universal translators, the future everywhere is AI, and America’s Cup racing is no different.

The power of intelligent machines is a vital area where Land Rover’s research and development division is adding considerable strength to Land Rover BAR’s challenge to win the America’s Cup – the challenge to bring home the world’s oldest international sporting trophy for the first time since it was originally contested in 1851.

Movie goers and readers of Michael Lewis’s Moneyball will know that big data has only recently made an impact on sports such as baseball and football; but Cup teams have been using sophisticated data analytics for a long while. The use of AI – and the subsidiary technology of machine learning – is just the latest tool brought to bear on a complex problem.


Machine learning evolved from the study of pattern recognition and computational learning theory in artificial intelligence (AI) *.  This is the same technology that allows Facebook to recognise faces in photos and suggest tagging, and makes voice commands possible on smartphones.

It differs from conventional computer technology because the machine is not explicitly programmed to execute a series of actions, rather, it’s programmed to learn.

Sailing is an almost perfect application for these technologies because there are so many variables and one fundamental unknown.

The team’s testing boats have a sensor array that measures over 300 variables from fibre optic strain measurement to six-axis accelerometer and gyroscopic motion sensors. Those are the variables, often recording a value many times a second, scooping vast quantities of information into a database.

Patterns

The unknown is the power source; the wind. The faster and harder the wind blows, the quicker the boat can go. Unlike a Formula One car – where the engine’s power output can be calibrated and measured very accurately and is very repeatable from test to test – the wind is constantly changing in speed and angle, and changing differently at different heights during each test. These micro changes are very difficult to measure, and it’s tough to quantify performance improvements when the power input is unknown.

This is where Land Rover’s machine-learning expertise comes in. The intelligent algorithms can sort through vast quantities of data and see patterns in the variables; patterns that could never be recognised by a human analyst. Those patterns allow the factors that are making a difference in performance to be recognised.

Artificial Intelligence: The power of machines

The man from Land Rover tasked with guiding the machines to this goal is Mauricio Muñoz, part of their Self Learning Car team. Jaguar Land Rover Advanced Engineering team has been adopting Big Data Analytics and Machine Learning into their processes and technologies to deliver unique customer features and support engineering decision making with real life data.

Jaguar Land Rover were the early adopter of Big Data and Machine Learning techniques within automotive, with the concept of the Self Learning Car - a vehicle that uses various data sources to understand the environment, driving patterns and features usage; it then removes daily driving chores and offers a completely personalized driving experience.

Originally from Ecuador, Muñoz’s parents had studied in Germany, and often spoke the language around the house. So when it came time to chose a college he decided that a German university would suit his desire for something both familiar and a little bit different – he’d already attended an American school while growing up in Quito, Ecuador.

Muñoz spent eight months learning the language in Germany, before applying to universities and being accepted to study computer science. He went on to do a Masters; a combination of software engineering (in Munich, Germany), and machine learning (at MIT in Boston). He was recruited directly from there to Jaguar Land Rover.

“At Jaguar Land Rover we are using machine-learning techniques to identify behavioural patterns, how people go about interacting with their cars, with the end goal of automating a lot of it,” explained Muñoz.

© Richard_Newton

“Machine learning is one of those buzzwords that gets thrown around a lot,” he added. "People get really excited about it and don’t really know what it is. My degree in computer science majored in computer vision, which is the discipline of teaching computers to see and to interpret video and images.

"One of the foundations of that is machine learning, so I’ve been working on it for a while. To me, machine learning means detecting patterns and using that knowledge to interpret something about the world.”

Hidden in the data 

Muñoz’s first task has been to analyse specific manoeuvres and straight line performance, looking for the optimal way to sail the boat. “In many contexts, machine learning only cares about one question; is there a pattern there, yes or no? If there is a pattern I want to use it. I don’t really care what the pattern is, I just want to be able to use that knowledge to my advantage.

"An example is Google's AlphaGo, which exploits patterns learned from professional Go players to beat the world's current best player. But developers at Google don't quite understand why that algorithm has chosen to behave the way it has; i.e. it has achieved its intended purpose (beating the other player) without saying why or how (making the patterns understandable to the developers). 

"The project at Land Rover BAR is a bit different, the focus is not so much whether the pattern is there — we know there are high performance tacks and gybes — but what is causing the pattern?”

The 34th America’s Cup established just how important it was to maintain stable flight on the foils through these manoeuvres, but what is it that makes the difference between staying airborne and crashing back into the waves?

It’s somewhere in the data, and Muñoz and his intelligent machines intend to find out.