Multi-sensor data fusion for autonomous vehicle situational awareness

Autonomous Vehicles (AVs) are expected to deliver a transformative impact on the long standing transport problems of capacity, congestion, safety, environmental impact and efficiency. The UK Connected Places Catapult estimates the global Connected Autonomous Vehicles (CAVs) based smart mobility market at £900 billion by 2025. Critical to successful operation of AVs is the capability for Situational Awareness (SA). Because of a lack of a single technology for SA, consideration must be given to the use of complementary multiple sensors including those for (1) PNT such as Global Navigation Satellite Systems (GNSS), Dead Reckoning sensors (e.g. Inertial Measurement Units, odometers, magnetometers) and signals of opportunity; (2) remote and environmental sensors such as cameras, LiDAR, RADAR and ultrasonic sensors, (3) Spatial databases, (4) algorithms to collate and interpret the multi-sensor data, and (5) powerful processors to execute the algorithms and plan a safe path forward in real-time. In addition to cost, security, and ethics, there are significant technical challenges. For example, the current SA systems are (1) expensive and do not the meet the required PNT accuracy, trustworthiness, continuity and service availability to support mission critical autonomous vehicle operations, (2) cannot provide the required high resolution maps in near real-time, (3) rely on weather-dependent sensors, such as LiDAR, and low resolution road maps for obstacle detection, environment sensing and navigation, and (4) the two groups of sensors (PNT and remote sensing) do not interact to provide the required navigation performance. To date, tests and ensuing successes have largely been demonstrated in good weather and using road maps with the required level of resolution, not currently widely available. Therefore, the objectives of this PhD research are to develop high performance SA models and algorithms that (a) better account for the effects of the weather effects and are self-contained and capable of delivering the levels of PNT accuracy, integrity, continuity and availability, (b) low cost, and (c) fuse all available sensor data to realise complete SA. Novel methods for characterising the performance of various sensors in physically complex city environments (i.e. heat mapping) will be developed and used together multi-objective optimisation algorithms to formulate and test the best sensor fusion system architecture for SA.
EPSRC Research Areas: Sensors and instrumentation, Infrastructure and urban systems