Computer systems which can learn from us and then help to carry out complex tasks may become a reality thanks to new research being carried out by the Computer Vision Group.
The research aims to tackle one of the long-standing problems in human-computer interaction — how to build computer systems which are able to understand and predict human behaviour and activity.
The research, part of a €3.2 million EU-funded project entitled ‘COGNITO’, will investigate the learning of human sensory-motor activities. The Bristol team, led by Dr Andrew Calway and Dr Walterio Mayol-Cuevas from the Computer Vision Group, believe the project will help to address one of the major stumbling blocks to computers becoming truly useful.
Dr Calway said: "I'm sure many people have experienced that moment when they have no idea what their computer is doing and it appears to have absolutely no idea what they are trying to do. Obviously this isn't a disaster when word processing, but it becomes a serious show stopper when we're trying to build systems that can genuinely help and interact with people.”
Using data captured from on-body sensor networks the project will investigate how to design systems that can observe and then learn how people do things. This knowledge can then be used to build systems which understand human actions and are able to offer help when required.
The Bristol team will carry out research into advanced computer vision algorithms to monitor human activities using on-body cameras. The work will form a central component in the project, for both the learning of actions and the use of that knowledge in assistive systems based on augmented reality technology.
A particular focus will be on learning skilled assembly and manipulation tasks that are often found in high-precision industries.
Dr Mayol-Cuevas added: "In Europe, where many industries from family-run to large corporations rely on highly-skilled people, systems that assist in rapid training and quality control are essential.”
Further information is available on the COGNITO project website.