Is it possible to develop a machine vision system that can think for itself? One that can recognize potential errors before they happen and never repeat the same mistakes? A group of scientists and engineers in the United Kingdom are hard at work developing advanced vision systems and artificial intelligence technologies that work together to address these issues and many others. Their work is likely to have wide-ranging impacts on factory automation, military defense systems and vehicle control. Under the umbrella "Project Reverb: Reverse Engineering the Vertebrate Brain," defense and aviation equipment supplier BAE Systems (www.baesystems.com) has joined forces with universities in Sheffield, Manchester, Bristol, Cambridge and Aberystwyth to recreate how the human eye and brain interact: "What we would like to do is develop systems that prepare and respond to the unexpected. We're looking to develop a vision system based on rapid decision-making-how the human eye focuses on objects and understanding what the eye is really looking at and how that is transmitted back to the brain and how the brain reacts-both from the point of deliberative decision-making to reactionary response," says Hector Figueiredo, group leader for the advanced information processing department at BAE Systems.
With assistance from researchers at the University of Manchester, the team has been using vision system technologies based on SCAMP-3 chip technology. SCAMPs are software-programmable vision chips that integrate an image sensor and processor array capable of performing more than one billion (a.k.a., giga) instructions per second. To their surprise, the chips weren't able to meet the operating requirements outlined: "It certainly is quick, but no where quick enough that I feel comfortable using it for full autonomous applications," Figueiredo says. The team is hard at work developing chip sets that perform computations even faster for improved reactive thinking performance, along with better imaging resolution to allow the machines to detect minute changes in color, shape, size and texture.
Figueiredo and his team have a deadline of 2010 to produce commercially feasible systems that will enable robots and other vehicles to operate autonomously using technologies developed under the Reverb program. He envisions these technologies revolutionizing the way manufacturers approach factory automation in the future. The team has begun testing robotic arms equipped with SCAMP chips, although the validation is still in its early stages. "I could foresee a time when the robots in the factory will be able to do a multitude of functions with one vision system, from assembly to final inspection before the part is delivered to the next station. These same systems will also be able to help predict and correct potential problems as they happen, without any user input," Figueiredo says.
The biggest remaining challenge is system integration-combining these complex technologies into a single system has proven to be daunting. "We're taking a systems approach to this project and not focusing on a single point solution," Figueiredo says. Mimicking the processing capabilities of the human brain is another challenge, requiring advanced systems that can pinpoint how the brain acts when it comes to vision and reactive thinking. Tools like MRIs, which provide detailed images of the brain, give insight into a mere fraction of what's happening in the brain. "Using the technology we have available to us is like looking at what's on Venus through a pair of binoculars," Figueiredo says. “We're making small steps and I wish we could go faster but I am confident we will be the ones who pull this together faster than anyone else."