QUT Centre for Robotics’ Dr Tobias Fischer, Professor Michael Milford, and PhD candidate Somayeh Hussaini have been awarded funding from Intel’s Neuromorphic Computing Lab for a project that will use Intel’s Loihi 2 neuromorphic research processor.
QUT said that based on the concepts of biological neural computation, neuromorphic computing employs brand-new algorithmic techniques that mimic how the human brain interacts with the outside world to produce capabilities that are more closely related to human cognition.
Spiking neural networks (SNNs), revolutionary computer models that resemble the brain’s neurons in our brains and simulate natural learning by dynamically re-mapping neural networks, are used by the chip to make judgments in response to patterns that have been learned over time.
Dr Fischer said algorithms for Simultaneous Localization and Mapping (SLAM) are at the foundation of autonomous mobile vehicles deployed in unfamiliar, GPS-deficient, or dynamic situations.
“Our project will focus on Visual Place Recognition, a particularly challenging part of SLAM that requires robust recognition and discrimination of hundreds of thousands of locations in different conditions. This will include ‘slow’ adaptation (urban vs rural, or day vs night) as well as ‘fast’ adaptation (sudden onset of rain, entering tunnels). Continuous adaptation is very challenging as it is impossible to cover all environment variations in the training data,” Dr Fischer stated.
According to Professor Milford, the QUT research team from the School of Electrical Engineering & Robotics was inspired by the animal realm to meet this challenge.
“Many animals display remarkable navigation capabilities, solving large-scale memory formation and recognition problems with unprecedented efficiency, flexibility, and robustness. New features of Intel’s Loihi-2 enable more complex neuronal models, making it well-positioned to implement the next generation of biologically inspired navigation and map formation algorithms that could surpass today’s state-of-the-art in the field,” Professor Milford added.
Dr Yulia Sandamirskaya, a research scientist at Intel Labs, Dr Garrick Orchard, a research scientist at Intel Labs, and Dr Andreas Wild, manager of software and algorithms at Intel Labs, have been working closely with the QUT research project team to develop algorithms.
“Neuromorphic computing opens new exciting new possibilities in computing, especially when it comes to robotics. Loihi 2 greatly improves the speed, programmability, and capacity of neuromorphic computing, which enables it to support more difficult optimization problems,” Director of Intel’s Neuromorphic Computing Lab Mike Davies said.
The prestigious Telluride Neuromorphic Cognition Engineering Workshop in Colorado featured the topic area “From Neuroscience Theory to Robotic Applications,” organised collaboratively by QUT and Intel Labs.