Machine Learning

Machine learning is a sub-sector of Artificial Intelligence where the machine is trained how to learn. It is an emerging research area as scientists are able to develop models quickly and analyse complex and big data faster.

Our research falls into these 4 areas:

Driver-machine integration

Our work around driver modelling and characterisation supports systems, ensuring the safe and timely transfer of control between autonomous and driver modes. Underpinning the emerging science of allocation of function is the development of machine learning. Our expertise in driver behaviour recognition and distraction detection has allowed us to create systems that recognise patterns of driver behaviour and identify abnormal activity. Personalisation through machine learning tailors driver behaviour monitoring for individual driver style and personality, improving reliability and safety.

Robust decision-making

We are developing a range of decision-making support tools – including optimisation algorithms, planning and scheduling tools, and anytime algorithms for real-time implementation-these systems ensure robust decision-making within an ever changing driving environment.

Advanced computing

Modern automotive embedded systems – spanning ABS, electronic stability control, traction control and automatic four-wheel drive – all rely on system on chip (SoC) technology. We have made significant advances in the development of SoC capable of interpreting image data and generating associated metadata. By developing computer vision and data interpretation in tandem with machine learning, we can create powerful systems to support autonomous situational awareness and robust decision-making. We are also exploring how SoC technology could support studies of driver behaviour and vehicle autonomy to underpin developments within allocation of function.

Enhanced mapping technology

Intelligent transport systems and services (ITSS) rely on high-quality real-time positioning data, in part generated by map-matching (MM) algorithms that ensure the data plots accurately onto the actual road network. Our Transport Studies Group has worked to enhance the reliability and accuracy of MM technology, and developed a number of knowledge-based intelligent MM algorithms that demonstrate greater integrity. We have developed and assessed a range of algorithms, using a variety of approaches including fuzzy logic, probability, genetic algorithm and Kalman filtering.

On-going research includes:

• the development of an optimisation technique to augment ITS services

• enhancing the algorithm by simultaneously, rather than sequentially, considering all uncertainties

• an objective comparative evaluation of existing integrity methods under the same circumstances to ascertain more robust evidence regarding algorithm performance

• traffic and safety assessments of IM

• advanced IM capabilities for partially and fully autonomous vehicles

Leading Group

Vision, AI, Autonomous and Human Centred Systems (VAAH)