I am Professor of Intelligent Systems in the Department of Computer Science, at Aberystwyth University. Most of my teaching career has been in Artificial Intelligence (AI), and my main research interests have concerned the application of AI to robotics.
My research work can be described in three phases. I started in the 1980s applying AI (rule-based systems, sensory reasoning) in industrial robot scenarios. The focus was error-recovery – how could a robot detect and recover from an error on its own? This led into the second phase when we realised that the robots need a model of the physical world of the objects that they were manipulating. This involved model-based reasoning, with applications in automated diagnosis. A spin-off was the discovery that modelling electric circuit theory was very amenable to our qualitative reasoning technique and we produced a novel method that formed the basis of a set of industrial software tools now in world-wide use in the automotive and aerospace sectors. During this period I was Co-founder Director of the European Network: MONET.
Since the turn of the century, the third phase, I have been working in Developmental Robotics; a new field that is very active and very exciting. Not enough attention is given to development, which is the key to human growth and learning. Fifty years of AI and robotics research has almost completely ignored human development and only taken inspiration from adult learning and adult behaviour. Yet human infants learn at fantastic rates and achieve so much. They gain control over complex sensory-motor systems, build up an understanding of three-dimensional space, learn to recognise subtle patterns, and start to communicate and interact with others.
So how can development be built into robots so that they may learn more effectively? Well, the source of much expertise and knowledge on infant development is to be found in Developmental Psychology. Psychologists study behaviour and we can draw on their theories and findings to create new mechanisms, computational models and algorithms that attempt to produce similar behaviour, and then lead on to novel robot learning techniques. I believe it is important to start at the beginning and so the data on post-natal and pre-verbal infants provides the grounding for initial sensory-motor experience upon which more advanced skills can be built.
My recent research grants have funded the implementation of experiments on various robots to support and progress this work. These include the DVL project (2002-06, EPSRC), the REVERB project (2005-10, EPSRC), the ROSSI project (2008-11, EC FP7) and the IM-CLeVeR project (2010-13, EC FP7).
My current project is the MoDeL Project (2015-18, EPSRC) entitled: “Developmental algorithms for robotics: Understanding the world of objects, interactions and tools”. This work is investigating mechanisms that drive robots to autonomously develop new behaviors and learn about novel events for which they have no prior experience. This involves intrinsic motivation, essential for true autonomous behavior, and we are exploring a play generator that produces behavior analogous to infant play with objects.
Scientific Advisory Panel: Dr Jacqueline Fagard, Paris Descartes University; Prof Merideth Gattis, Cardiff University; Dr Giorgio Metta, IIT, Genova, Italy; Dr Keith Mowbray, QinetiQ, UK; Prof. Kevin O’Regan, Paris Descartes University; Dr B. Stjerne Thomsen, The Lego Foundation, Denmark; Dr David Whitebread, Cambridge University, UK.
For a demonstration of longitudinal development on the iCub humanoid robot, see the video "Time Lapse learning"