Researchers from Oxford’s Department of Physics are demonstrating a departure from traditional computer chips by using water waves in a metal container to process information and control a robotic vehicle in real-time. Led by Dr. Safeer Chenattukuzhiyil, the team drew inspiration from the brain’s collective dynamics and a framework called reservoir computing, which requires training only at the output.
“One of the most exciting aspects of this project is that we are bringing together two worlds: fundamental physics and autonomous robotics,” Dr. Safeer comments. The work, published in Nature Communications, points towards compact, energy-efficient hardware for future autonomous machines and addresses limitations in current AI systems reliant on remote data centers.
Water Wave System Demonstrates Real-Time Obstacle Recognition
The water-wave computing system achieved 100% accuracy in robotic obstacle-recognition tasks after a training period, demonstrating a novel approach to sensory data processing. Researchers converted information about a robot’s surroundings into water waves within a circular metal container, then observed how wave interactions formed interference patterns that signaled different obstacle scenarios. An LED illuminated these patterns on the water’s surface, while a camera captured the projections for direct visualization of the physical computation, allowing the system to “learn” to identify obstacles.
This wave-based hardware then directly controlled a physical robotic car, with the generated wave patterns dictating its movements and enabling autonomous obstacle avoidance; the team also implemented an event-driven approach, triggering computation only upon detecting environmental changes. This minimized unnecessary processing, resulting in a more energy-efficient and responsive control system for the robot, a departure from the continuous operation of many conventional AI systems.
These simulations revealed that a spin-wave reservoir measuring just one micrometre in diameter could replicate the same robotic recognition tasks performed by the larger water-wave system. This suggests a pathway toward compact, energy-efficient neuromorphic hardware, potentially overcoming limitations of current GPU-based AI processing and enabling truly autonomous machines capable of real-time responsiveness and operation in complex environments. The research, published in Nature Communications, points towards a future where physical properties themselves become the basis for intelligent computation.
Oxford University Innovation Supports Wave-Based Computing Patent Application
Oxford University Innovation is now actively supporting a patent application for computing hardware that uses physical waves to control robotic systems, a step intended to move the technology beyond laboratory demonstrations. The University’s innovation partner, OUI, is assessing the commercial viability of the wave-based approach and identifying potential applications for the novel computing method, signaling a commitment to translating research into practical use.
This protection of intellectual property follows experiments demonstrating up to 100% accuracy in robotic obstacle recognition using a system that bypasses traditional processing units like GPUs. The research, led by Dr Safeer Chenattukuzhiyil, contrasts with conventional artificial intelligence systems that demand extensive training of entire neural networks, potentially offering substantial gains in energy efficiency and responsiveness, particularly for applications like autonomous robotics.
The system’s design draws inspiration from the collective dynamics observed in biological brains, where complex behaviors emerge from interacting neurons. Professor Thorsten Hesjedal, a co-author of the study, emphasized the simplicity of the experimental setup, stating, “For us, this was a reminder that impactful research does not always have to begin with an expensive or complicated experimental system.” The experiment involved converting information about a robot’s surroundings into wave patterns within a contained system, then using those patterns to directly control the robot’s movements, a method that makes the physical computation visible through projected interference patterns captured by a camera.
The resulting data, detailed in the Nature Communications article titled “Autonomous robotic operation controlled by wave-based neuromorphic hardware,” authored by J Zohar, D Pinna, G van der Laan, T Hesjedal, and CK Safeer, establishes a foundation for future development and application of this wave-based computing paradigm.
See today’s quantum computing news on Quantum Zeitgeist for the latest breakthroughs in qubits, hardware, algorithms, and industry deals.
