Eye, robot

2 min read

Computer models of human visual responses will help robots get their priorities right.

Multi-tasking robots capable of stopping what they are doing and responding to events in the corner of their ‘eye’ are being developed in a project involving

BAE Systems


The project, which began this week, is aimed at developing technologies to create robots with the ability to switch between tasks in the same way as animals and humans.

Researchers at BAE and UK universities including Sheffield, Manchester, and Cambridge are attempting to unravel the mechanisms the vertebrate brain uses to multi-task seemingly effortlessly. This research will be used to build computational models that can control the behaviour of a wide range of robots.

These could include devices for going into dangerous, hostile or inaccessible environments, monitoring the environment and human performance for signs of danger or assisting the disabled, aged or infirm. Each of these tasks calls for a robot that can adapt to and learn from unpredictable circumstances, said Dr Kevin Gurney, the project leader and lecturer in computational neuroscience at Sheffield.


‘It’s about attention and how you decide what to look at next. If you are preoccupied with a task you may be doing it well, but if someone comes along and attacks you with a hammer you aren’t going to live very long,’ he said.

BAE believes the technology could be used in devices such as its laser-guided Crawler, developed with Airbus, for carrying out tasks such as machining and inspection of aircraft parts. As such robots could be working in groups and in the same environment as humans, an ability to interact with others and plan their actions could improve safety and efficiency.

The £1.9m EPSRC-funded project, called Reverb (reverse-engineering the vertebrate brain), builds on work the Sheffield researchers have done on the region of the brain known as the basal ganglia, which is believed to act like a switch in deciding how to prioritise actions. The robot control system will use a central selection mechanism based on this process.

Subject to internal approval by BAE, the team hopes to enter Darpa’s Grand Challenge, the unmanned vehicle race that last year ended in fiasco when the most successful robot managed to travel only seven of the race’s 150 miles.

Dr Hector Figueiredo, group leader for mimetics at BAE’s Advanced Technology Centre in Filton, said he believes his team has a strong understanding of the technologies required to succeed.

‘If a person drove a car from one end of the course to the other, the brain is fantastic in the way it responds to the obstacles, the changing environment and the road surface. It can take all that information on board, fuse it together and make decisions quickly.

‘While we can already do that in robots using conventional techniques, I’m hoping that by taking a brain-inspired approach we will be able to do it in a more rapid and robust manner,’ he said.


Meanwhile the researchers are planning experiments in which a pan and tilt camera, acting as the eyes of the system, will be integrated with a robotic arm. The camera will look for interesting objects on a table and the arm will try to pick these up and put them to one side. The system will try to learn which are useful and which are not. But at the same time the system must be able to cope with interruptions by stimuli in the periphery of its vision.

The team will also use BAE-built mobile robots which will roam around cluttered environments where elements randomly appear and disappear. The cameras will focus on features such as moving objects, and the robots will take appropriate action.

The vision system will be built around a high-resolution ‘foveal’ camera and a lower resolution peripheral camera, which is similar to the way the retina is built, said Dr Piotr Dudek, lecturer in integrated circuit engineering at Manchester University.

To ensure the system operates in real-time the team will use custom-built parallel processing hardware based on vision chips developed by Dudek. This technology integrates image sensing and processing on to a single silicon chip, he said.