Classical and Quantum Physical Reservoir Computing for Onboard Artificial Intelligence Systems: A Perspective

TL;DR

The paper tackles the power bottleneck of onboard AI in autonomous platforms and surveys energy-efficient alternatives based on reservoir computing, including classical physical RC using environmental nonlinearities (turbulence, water waves, vibrations) and quantum RC (spin networks, Kerr oscillators, measurement-driven dynamics). It highlights architectures such as water-wave RC for UAV/ROV propulsion, whisker-sensor RC for UGV terrain adaptation, and diverse QRC platforms with potential for low training costs and compact hardware. Key contributions include synthesizing experimental and theoretical work across fluids, acoustics, mechanics, and quantum systems, outlining practical onboard implementations, and discussing measurement protocols to enhance speed and robustness. The work underscores the practical impact of energy-efficient RC in extending range and capability of autonomous vehicles and identifies a pathway toward green AI through interdisciplinary collaboration between physics, engineering, and quantum technologies.

Abstract

Artificial intelligence (AI) systems of autonomous systems such as drones, robots and self-driving cars may consume up to 50% of total power available onboard, thereby limiting the vehicle's range of functions and considerably reducing the distance the vehicle can travel on a single charge. Next-generation onboard AI systems need an even higher power since they collect and process even larger amounts of data in real time. This problem cannot be solved using the traditional computing devices since they become more and more power-consuming. In this review article, we discuss the perspectives of development of onboard neuromorphic computers that mimic the operation of a biological brain using nonlinear-dynamical properties of natural physical environments surrounding autonomous vehicles. Previous research also demonstrated that quantum neuromorphic processors (QNPs) can conduct computations with the efficiency of a standard computer while consuming less than 1% of the onboard battery power. Since QNPs is a semi-classical technology, their technical simplicity and low, compared with quantum computers, cost make them ideally suitable for application in autonomous AI system. Providing a perspective view on the future progress in unconventional physical reservoir computing and surveying the outcomes of more than 200 interdisciplinary research works, this article will be of interest to a broad readership, including both students and experts in the fields of physics, engineering, quantum technologies and computing.

Figures (10)

• Figure S1: (a--c) Schematic illustration of the physical processes---turbulence, water waves and vibrations caused by surface roughness---that onboard AI systems can employ as a means of energy-efficient computation. The so-envisioned AI systems can be used in UGVs, UAVs and ROVs as well as human-operated aeroplanes, boats and cars. (d) Identified as a separate category, quantum-mechanical physical systems can be used as onboard AI systems. Superior computational performance and low power consumption of quantum systems compared with traditional computers render them especially useful for applications in lightweight and long-range drones.
• Figure S2: Schematic representation of (a) a traditional algorithmic RC system and (b) a computational system with a physical reservoir constructed using the physical effects that take place in the environment that surrounds moving vehicles. The mathematical meaning of the vectors and matrices mentioned in this figure is explained in the main text.
• Figure S3: (a) Photographs of the prototype of ROV designed to test AI systems that employ water waves and other disturbances caused by the motion of the drone as a means of computation. (b) Schematic illustration of the difference between the wave patterns produced by the ROV that has received the left, forward and right commands from the operator.
• Figure S4: (a) Illustration of the vortex shedding taking place when a fluid such as air or water flows past a cylinder. As theoretically shown in Ref. Got21, modulating the flow velocity and monitoring the vortex dynamics using a set of virtual sensors one can create an efficient physical RC system. An experimental implementation of this computational approach was discussed in Ref. Vin23_1. (b) Photograph of vortices and other water flow effects created by the ROV in a lab setting.
• Figure S5: Sketch of an ROV, bubbles created by its propellers and highly nonlinear acoustic waves emitted by them. As discussed in the main text, the nonlinear dynamics of these physical processes opens up novel opportunities for physical reservoir computing.