Robotics Using Wave-Based Neuromorphic Hardware

Equipping robots with physical, brain-inspired computing hardware is a key step towards achieving greater autonomy in intelligent machines. However, the practical realization of neuromorphic systems for robotics remains a significant challenge. Conventional robotic platforms increasingly rely on energy-intensive deep neural networks and remote computational infrastructure, which can limit their operation in dynamic, resource-constrained, or remote environments and motivate the development of dedicated neuromorphic hardware [1].

In our recent work, we demonstrated autonomous robotic control enabled by wave-based neuromorphic computing hardware [2,3]. Using an experimentally realized water-wave reservoir computer, we show real-time processing of robotic sensory information for obstacle recognition and closed-loop navigation, with the dynamics of the physical wave system directly performing the required computation. We further extend the same computational framework to electrically excited spin waves using micromagnetic simulations, demonstrating its potential for nanoscale solid-state neuromorphic processors operating at GHz frequencies. Together, these results establish a general approach for exploiting interacting wave dynamics to perform intelligent information processing directly in physical hardware. Our work positions wave computing as an alternative pathway to conventional digital neuromorphic computing, offering a potential route towards energy-efficient hardware for autonomous robotics, edge intelligence, and other artificial intelligence applications.

Y. Sandamirskaya, M. Kaboli and J. Conradt, Science Robotics., Vol. 7, p.eabl8419 (2022) J. Zohar, D. Pinna, G. van der Laan, T. Hesjedal, and C.K. Safeer, Nature Communications (accepted) C.K. Safeer, J. Zohar, D. Pinna, G. van Der Laan, T. Hesjedal, Patent: GB2513667