The School of Engineering of the University of Glasgow is seeking a highly motivated graduate to undertake an exciting 3.5-year PhD project entitled
Application for this scholarship is made by using the online system at the following link:
(please select from list)
Please note that this application is to gain admission to our PGR programme, and an offer of admission may be issued before a decision on this Scholarship is made. Candidates applying for this Scholarship will most likely have an interview/discussion with the supervisor before any decision is made.
Autonomous systems are reshaping the world by greatly extending human capabilities in traditionally unreachable situations. Home service robots, search-and-rescue robots and self-driving vehicles are significantly changing the modern society. We expect that autonomous systems are responsible to their actions. Furthermore, we expect robotics can understand morality, ethics and compassion. To this end, autonomous systems and humans need to effectively communicate with each other and understand their goals. This requires autonomous systems to interact with humans, themselves and environments efficiently.
Traditionally, autonomous systems are represented by rule-based systems or machine learning systems. However, rule-based systems are brittle because slight changes to their working assumptions require the rewriting of rules. Machine learning, in particular deep learning, are not scientifically transparent. The lack of transparency indicates that the operations cannot be proven in a certain way. In the real world, autonomous systems have to face challenges including uncertainties, interactions, complex dynamics and diverse environments. A probabilistic approach is usually used to robustly address the challenges. However, probabilities cannot distinguish between cause and effect. Therefore, the challenges may be sorted improperly. For example, probabilities from statistics cannot express simple statements such as “mud does not cause rain”. Autonomous systems that robustly solve real-world tasks need to reuse and repurpose their knowledge and skills in novel scenarios, which is the power of machine reasoning. This research aims to fundamentally promote the autonomy level by reforming representation and decision making with machine reasoning. In other words, autonomous systems are expected to ‘learn from purposes’.
Autonomous systems should be able to solely accomplish missions. Machine reasoning will be a powerful and essential tool to realise this task. In this project, the causal models supporting explanation and understanding should be built. Then, intuitive physical theories and cognition science should be combined to enrich the learnt knowledge and skills. Afterwards, decision making mechanisms driven by causal reasoning would be developed. The objectives in decision making include: (1) Identifying representation models of variables and their relations in complex environments; (2) Incorporating physics and time-series data into the model; and (3) Learning the representation model dynamically. Autonomous systems would be allowed to make more adaptable decisions facing new situations and new tasks.