The eyes of the future
by Rodney Magazinovic
In
the animal kingdom, vision is one of the most powerful tools animals
have to navigate their three dimensional environment in search of food,
to find a mate and to avoid predators. And it is that acute visual
capability that UniSA researchers are trying to replicate to develop the
next generation of mobile robots.
Scientists now believe that when animals travel down a path they do not memorise all of the features of their surroundings but rather, only the salient ones. The recognition of these prominent features while negotiating an environment serves as a navigational reference that gives animals a perceptual map of a three dimensional environment. Referred to as topological navigation, it is that animal ability that has been essential in developing autonomous mobile robots that can recognise visual landmarks.
Most conventional autonomous robots have been built to rely less on vision to navigate because it is computationally intensive. This new focus on the visual environment makes the UniSA research project unique, according to robotics PhD candidate Quoc Do.
Do has developed a robot that comprises a navigating vehicle and an off-board vision system that is mounted on the framework of a remotely controlled car. It communicates with a remote PC via wireless video and data links. Do says the robot can navigate through an environment by recognising visual landmarks in both clean and cluttered backgrounds.
“The robot will only need to recognise the significant features placed at the most critical points along a certain pathway,” he said. “By using this information as a reference point it can adapt and change its path if necessary.”
The robot has been able to successfully navigate both indoor and outdoor environments.
A research collaboration between UniSA’s Knowledge-Based Intelligent Engineering System Centre at Mawson Lakes and the Defence Science and Technology Organisation (DSTO), the “seeing” robot has enormous potential.
Research supervisor for the project, DSTO’s Dr Peter Lozo said UniSA was selected as a collaborative partner because of its experience in visual image processing and neural networks.
“This research has obvious implications for the defence industry – in navigating dangerous terrain, working on landmine clean-up and similar uses,” Dr Lozo said.
“But there is also enormous scope in applying the technology in a range
of ways from the simplest domestic applications such as autonomous vacuum
cleaners and lawnmowers, to improved mining and manufacturing systems.”