Researchers at a Queensland University of Technology (QUT) analysis group have drawn inspiration from the neural networks of insects and animals to develop more energy-efficient robotic navigation systems, mimicking nature’s remarkable ability to optimize movement while minimizing energy expenditure.
Led by postdoctoral analysis fellow Somayeh Hussaini, in collaboration with Professor Michael Milford and Dr. Tobias Fischer of Queensland University of Technology’s Centre for Robotics, a novel place recognition algorithm was developed and published in a leading journal, supported by Intel Corporation, which utilizes Spiking Neural Networks (SNNs) to achieve improved performance.
“Synthetic neural networks, or SNNs, simulate the way organic brains process information by employing brief, discrete signals reminiscent of the communication between neurons in animal brains,” she explained.
“These networks prove particularly adept at leveraging neuromorphic hardware – innovative, brain-inspired computing systems that mimic the efficiency of natural neural networks – allowing for accelerated processing and drastically reduced energy expenditure.”
While significant advancements have been made in robotics, stylish robots still face the challenge of successfully navigating and operating in complex, unfamiliar settings. While they often rely on AI-generated navigation methods that demand crucial computational and energy resources from their training regimens.
“Dr. Fischer observed that animals possess an impressive ability to navigate complex, ever-changing environments with remarkable efficiency and resilience.”
“This research represents a significant stride towards the development of biologically inspired navigation methods, which may ultimately rival or even outperform conventional approaches.”
Developed by the Queensland University of Technology (QUT) group, the system leverages small neural network modules to identify specific locations from images. These modules have been integrated into an ensemble, a collection of multiple spiking networks, to form a scalable navigation system capable of navigating vast environments through studying.
“By leveraging sequences of images rather than individual frames, Professor Milford’s team achieved a 41% boost in place recognition accuracy, allowing the system to effectively adapt to changes in lighting conditions, seasonal variations, and weather patterns over time.”
The system was successfully showcased on a resource-limited robotic platform, effectively proving its viability in practical scenarios where energy efficiency is paramount.
This pioneering research has the potential to clear a path for the development of eco-friendly and reliable navigation strategies for autonomous robots operating in low-power scenarios. “As she underscored, exciting breakthroughs in fields such as space exploration and disaster relief hinge on the strategic optimization of energy efficiency and swift response times.”
