A driver's actions impact on the safety and stability of the vehicle. If we can accurately predict future driver inputs (steering wheel angle, brake/throttle pedal position etc.), then we can start to make plans for the vehicle's future trajectory.
For example, trucks are limited how quickly they can go around a corner by the vehicle's maximum lateral acceleration. If this is exceeded, the truck rolls over. By estimating the likely future inputs from the driver, we can predict rollover events in advance and take evasive action (actuate the brakes, limit steer angle) to reduce the risk of a rollover.
In this project we'll investigate methods for doing this prediction, with the intention of mapping out a set of future states and their likely probabilities. This will be done with machine learning and likely use neural networks.
Machine learning | Driver behaviour | Neural networks | Accident prevention | Behaviour prediction
This project has been suggested by Volvo Trucks Sweden. We will be reporting our findings back to them at the end of the project, as well as meeting with them during the project to update them on our progress and ask for feedback.
We have gigabytes of driver data to analyse, covering millions of miles of driving. This is a unique opportunity to work with real-world data sets in collaboration with an industry partner.
We will meet once a week to discuss progress and decide the next steps.