McLaren believes in extracting maximum insight from data, be it big or small, and it’s this insight that we are using to solve industry challenges.
Our Engineering teams are based in London and Singapore, and work closely together to tackle challenging projects. We sit in the sweet spot where simulation modelling and data science intersect, and we believe our experience with combining these two approaches in Formula 1 gives us a unique perspective on the increasingly complex data challenges that we are seeing today.
At the heart of all our work is helping people to make better decisions in uncertain environments, whether it concerns the operation of a race-car, the performance or health of an individual, or the optimisation of an industrial process or airport. Creating a model of the system in question allows us to understand its behaviour, to test out different ‘what-if’ scenarios, and to ultimately learn how to maximise its performance, either through human intervention or through automated control.
There are different methods for creating such a model: you can collect lots of data and use statistical modelling and machine learning, or you can start to fundamentally understand and then codify the behaviour of the system. A data-driven approach is more scalable, whereas a mechanics-driven approach better extrapolates to new scenarios.
Within our team, we span both approaches – indeed we’re most interested in where these two approaches can come together: where we can exploit fundamental system understanding to complement machine learning and where machine learning can utilise data to improve traditional simulation and design methods. Our approach also represents our desire to understand how and why systems behave the way that they do.
An example of how we combine machine learning and physical modelling is in adaptive vehicle control. There are well-established physics-based models of vehicle components, but the parameterisation of those models may be poorly understood or subject to change, for example the friction coefficient of a tyre. Our approach enables us to refine our model by learning its parameters from data. Furthermore we are able to quantify the uncertainty associated with the model in a scalable way using modern machine learning techniques and tools.
This unified approach is crucial to our industry focus areas: healthcare, transport, automotive and motorsport. Human physiology can be simulated using generic models, but as healthcare becomes more personalised, so will our approaches need to be. Population health data will give us huge insight into care needs but these are bounded by regulatory constraints. Logistical systems are rigidly defined but plagued with uncertainty which needs to be characterized to ensure that optimal plans are robust. Autonomous systems will be subject to physical and operational constraints that are already well understood.
Immersive simulation and decision-support environments require us to characterise the human-machine interaction, not just the two in isolation.