Researchers at Queensland University of Technology have achieved a major breakthrough, developing a novel robotic navigation system that mimics human brain neural processes, using energy less than 10% of traditional systems, bringing new advancements to robotics technology.
The related research results were published in the journal Science Robotics, with the paper titled "A Compact Neuromorphic System for Ultra-low-power, On-device Robot Localization." The research team consists of neuroscientist Dr. Adam Hines, Professor Michael Milford and Dr. Tobias Fischer from the Queensland University of Technology Robotics Centre and the School of Electrical Engineering and Robotics. They conducted this study using neuromorphic computing systems.
To run the neuromorphic system, the research team designed specialized algorithms that enable it to learn like humans, processing information in the form of electrical spikes, similar to the signals used by real neurons. Dr. Hines pointed out that energy constraints are a major challenge facing real-world robotics technology, especially in fields such as search and rescue, space exploration, and underwater navigation. By using neuromorphic computing, the new system reduces the energy demand for visual localization by 99%, allowing robots to operate longer on limited power sources and cover greater distances.
In the study, the LENS (Location Encoding based on Neuromorphic Systems) system developed by the team performed exceptionally. The system can identify locations over an 8-kilometer journey using only 180KB of storage space, nearly 300 times less than other systems. LENS integrates brain-like spiking neural networks, special cameras that only respond to motion (event cameras), and low-power chips into a compact robot.
Dr. Hines stated that the system demonstrates how neuromorphic computing can achieve real-time, energy-efficient localization tracking for robots, opening new possibilities for low-power navigation technology. Lower energy consumption enables remote-controlled robots to explore for longer periods and over greater distances, and the system allows robots to quickly and energy-efficiently self-localize using only visual information.
ARC DECRA Fellow Dr. Fischer introduced that the key innovation of the LENS system is a new algorithm that leverages two promising bio-inspired hardware types: perception via event cameras and computation via neuromorphic chips. Event cameras do not capture complete scene images but continuously sense changes and motion every microsecond, detecting brightness changes in each pixel, closely replicating the way the human eye and brain process visual information.
Michael Milford, Director of the Queensland University of Technology Robotics Centre, said that this research represents a key research theme of the centre. Influential robotics technology not only requires breakthrough research but also completion of translation work to ensure it meets end-user expectations and requirements. This study is an excellent example of commitment to energy-efficient robotic systems, providing the performance and durability needed for end-users to enable robots to function effectively in application domains.
