A car fitted with a navigation system inspired by a rat鈥檚 brain has successfully mapped a 66-kilometre road network in a suburb of Brisbane, Australia. Built by Michael Milford and Gordon Wyeth at the University of Queensland in St Lucia, the system is designed to give future autonomous robotic vehicles the ability to explore and map their environment.
At its heart is a neural network that mimics the way a rat makes a mental map of an area as it scurries through it 鈥� a concept known as simultaneous localisation and mapping. Specialised neurons in a rat鈥檚 brain called 鈥減lace鈥�, 鈥渉ead-direction鈥� and 鈥済rid鈥� tell the animal what direction it is facing and if the environment is a familiar one. Neural networks inspired by rat brains have previously been used to guide robots around simple lab mazes, but Milford and Wyeth鈥檚 is the first to mimic the way grid neurons cope with an unfamiliar environment.
Running on a computer with a webcam feeding it video as the car drives around, the neural network maps an area by looking for landmarks such as junctions, trees and buildings, and recording the direction it is facing when it sees them. When it revisits an area and recognises a slightly altered view of a familiar landmark, it looks back over other landmarks it has recently passed to confirm its location, and updates its map accordingly. 鈥淚t works out where the car is in the suburb even when the visual information it鈥檚 getting is very ambiguous or confusing,鈥� says Milford.
Advertisement
A report on Milford and Wyeth鈥檚 system will appear in IEEE Transactions in Robotics later this month.
Robots 鈥� Learn more about the robotics revolution in our continually updated .
Cars and Motoring 鈥� Learn more about the latest technologies in our comprehensive .
