C-Lab: Cognitive Systems
C-Lab focuses on research broadly termed Cognitive Vision but encompasses machine learning, AI, Cognitive Systems and Robtics. The definition of a Cognitive System is one that has similar cognitive abilities to biological systems, specifically exhibiting robustness to unpredictable scenarios, adaptability and the ability to learn and/or reason about the world. Rather than being engineered from first principles to tackle a specific problem a cognitive approach attempts to build systems that have their own ability to learn and adapt to new events allowing flexibility, generality and robustness to be introduced to the system.
Example projects below demonstrate how weak supervision can be used along with embodiment to produce robots that are capable of both seeing and reasoning about the world.
Weak supervision in the form of linguistic annotation also plays a major role in learning e.g. see its use in Human Motion Analysis or in learning Sign from subtitles.
Work in the area of Visual Surveillance has developed automated approaches that can track multiple people across wide are surveillance systems without any prior knowledge of the scene or camera placement using people themselves as the calibration object and adapting models over time as new evidence becomes available.
In Video Mining we are developing new approaches to efficiently mine information from large corpus of broadcast and online content. By combining these learning approaches with new modes of Visual Interaction we allow the user to weakly supervise the learning process and become part of the cognitive learning framework.
Projects
V-lab has strong collaborations across Europe and has participated in numerous EU and EPSRC collaborative projects. Projects include:
COSPAL COgnitiveSystems using Perception-Action Learning
The COSPAL project aimed to build an artificial cognitive system to allow an autonomous robot equipped with a vision system to learn from first principles how to solve a childrens shape sorting puzzle. The architecture, based on the concept of modelling affordances in terms of percepts and associated actions used imitation and exploration to automatically learn, control, recognise and solve puzzles. For further information see the project website
URUS Ubiquitous Networking Robotics in Urban Settings
The URUS project focused on designing a network of robots that could cooperatively interact with humans and the environment. Specific tasks were those of assistance, transportation of goods, and surveillance in urban areas. One of the major objectives of the project was to design and develop a cognitive network robot architecture that integrates cooperating urban robots, intelligent sensors, intelligent devices and communications. For further information see the project website.
DIPLECS Dynamic Interactive Perception-action LEarning in Cognitive Systems
The DIPLECS project aimed to design an Artificial Cognitive System capable of learning and adapting to respond in the everyday situations humans take for granted. The primary demonstration of its capability was in the context of assistance and advice to the driver of a car. The system will learn by watching humans, how they act and react while driving, building models of their behaviour and predicting what a driver would do when presented with a specific driving scenario. The end goal of which is to provide a flexible cognitive system architecture demonstrated within the domain of a driver assistance system, thus potentially increasing future road safety. For further information see the project website.
People
Prof Richard Bowden
Prof Josef Kittler
Prof J Illingworth
Dr Bill Christmas
Dr David Windridge
Dr Eng Jon Ong
Dr Andrew Gilbert
Dr Helen Cooper
Dr Nicholas Pugeault
