Reservoir computing from collective dynamics of active colloidal oscillators

Veit-Lorenz Heuthe, Lukas Seemann, Samuel Tovey, Clemens Bechinger · 2026

Read Reservoir computing from collective dynamics of active colloidal oscillators by Veit-Lorenz Heuthe, Lukas Seemann, Samuel Tovey, Clemens Bechinger (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-05767.txt.

Introduction

Veit-Lorenz Heuthe and 3 co-authors turned hundreds of swimming colloidal particles into a computer in a 2026 demonstration of this idea. Reservoir computing feeds inputs into a complex nonlinear system and trains only a simple readout, saving the energy cost of training a full neural network. Most physical reservoirs cannot tune their internal interactions.

Here, hydrodynamically coupled active colloidal oscillators form a fully parallel system. Their coupling strength and memory time can be adjusted in place during operation. From single readouts, the system predicted chaotic time series accurately without time-multiplexing tricks. It also detected subtle hidden anomalies that changed a signal's timing correlations while leaving its ordinary statistics untouched, suggesting uses in smart sensing. Only the last, simple layer is trained. The swarm of particles does the hard work on its own, driven by the input signal.

Catalog record

CREATORVeit-Lorenz Heuthe, Lukas Seemann, Samuel Tovey, Clemens Bechinger
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE41 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-05767

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Reservoir computing from collective dynamics of active colloidal oscillators Veit-Lorenz Heuthe, Lukas Seemann, Samuel Tovey, Clemens Bechinger Abstract Physical reservoir computing is a computational framework that offers an energy- and computation-efficient alternative to conventional training of neural networks. In reservoir computing, input signals are mapped into the high-dimensional dynamics of a nonlinear system, and only a simple readout layer is trained. In most physical implementations, the interactions that give rise to the dynamics cannot be tuned directly and high dimensionality is typically achieved through time-multiplexing, which can limit flexibility and efficiency. Here we introduce a reservoir composed of hundreds of hydrodynamically coupled active colloidal oscillators forming a fully parallel physical reservoir and whose coupling strength and fading-memory time can be tuned in situ. The collective dynamics of the active oscillators allow accurate predictions of chaotic time series from single reservoir readouts without time-multiplexing. We further demonstrate real-time detection of subtle hidden anomalies that preserve all instantaneous statistical properties of the signal yet disrupt its underlying temporal correlations. These results establish interacting active colloids as a reconfigurable platform for physical computation and edge-integrated intelligent sensing for model-free detection of irregularities in complex time signals. Figure 1: Active colloidal oscillators. a A colloidal particle (grey) with position \mathbf{r}_{i} is activated by a focused laser beam (green), inducing self-propulsion with velocity \mathbf{v}_{i} . b Active colloid steered toward a target position \mathbf{q}_{i} (red). The propulsion direction \mathbf{v} (solid arrow) deviates from the target direction (dashed line) due to a time delay between particle detection and laser positioning. c This misalignment induces an orbiting motion of the active colloid around the target position (blue arrow). d Experimental snapshot of 400 ACOs with target positions arranged on a hexagonal lattice of spacing L . e,f Trajectories (blue) of a single particle for damping parameters \Gamma = 0 and 1.5\mathrm{\mu m} . The gray area corresponds to the optical image of the particle (dashed circle). g,h Section of the oscillator array shown in D , with ACOs colored by their angular momentum h_{i} for \Gamma=0 and 1.5\mathrm{\mu m} . i Experimentally measured probability distributions of \left|h_{i}\right|=\left|\mathbf{v}_{i}\times(\mathbf{r}_{i}-\mathbf{q}_{i})\right| for different values of \Gamma . Introduction The rapid expansion of artificial intelligence has intensified the demand for energy-efficient computing architectures. Contemporary machine-learning approaches, particularly deep neural networks, rely on optimizing millions of parameters and therefore require substantial computational and energy costs. As data volumes and task complexity continue to grow, alternative paradigms that perform computation directly within physical substrates (in materia) are attracting increasing interest. Reservoir computing (RC) offers such a route: instead of training all internal connections of a neural network, RC exploits the intrinsic dynamics of a nonlinear system—–the reservoir—–to transform input signals into a high-dimensional state space, while learning is confined to a simple linear readout. Physical reservoirs naturally provide nonlinearity and intrinsic memory, enabling compute-in-memory architectures that bypass the von Neumann bottleneck [ 20 , 12 ] . RC has been explored on a wide range of physical platforms, including optical feedback networks [ 4 , 33 ] , memristor arrays [ 11 ] , magnetic textures [ 2 ] , quantum devices [ 28 , 29 , 35 ] , and chemical or electrochemical systems [ 10 , 1 ] . These systems are particularly well suited for processing temporal signals with strong dependence on their history, enabling applications such as forecasting chaotic dynamics, interpolating missing data, and detecting anomalies in fields ranging from physiology and seismology to climate science and finance [ 21 , 31 , 6 , 37 , 34 , 42 , 30 ] . However, in most existing physical reservoirs, internal interactions are fixed or only weakly tunable, and high effective dimensionality is often achieved through time-multiplexing rather than by exploiting intrinsically rich reservoir dynamics [ 9 ] . This limits true parallel processing, restricts access to optimal regimes that balance nonlinearity and fading memory, and hampers systematic optimization, since modifying interaction strength or network topology typically requires the fabrication of new devices. These limitations motivate the search for physical reservoirs whose computational richness arises intrinsically from genuine many-body dynamics and whose interactions can be reconfigured in situ. Active matter, i.e., ensembles of self-driven colloidal particles with highly nonlinear and long-range interactions—provides a natural realization of such systems [ 14 , 27 , 13 ] . Although recent work has shown that single synthetic active particles with delayed feedback can serve as nonlinear physical reservoir nodes [ 39 ] , it remains an open challenge to realize reconfigurable reservoirs in which computational function emerges from collective, spatially extended dynamics of interacting units. Harnessing such many-body dynamics promises intrinsically high dimensionality, true parallelism, and computational capabilities that are inaccessible in single-node or time-multiplexed architectures. Here we demonstrate a physical reservoir realized as a hexagonal lattice of hundreds of hydrodynamically coupled active colloidal oscillators (ACOs). The platform offers full experimental control over coupling strength, lattice geometry, and signal injection, enabling systematic exploration and optimization of reservoir performance. Computation is encoded in the collective spatiotemporal dynamics of the array and proceeds in parallel across hundreds of interacting degrees of freedom, without reliance on time-multiplexing or virtual node construction. These collective dynamics enable accurate forecasting of chaotic time series and, crucially, real-time detection of hidden anomalies within chaotic signals. Together, our results establish colloidal active-matter reservoirs as a versatile experimental platform for edge-integrated intelligent sensing and processing complex temporal data. Active colloidal oscillator arrays Figure 2: Response of reservoir of 400 ACOs to an input signal. a-d Relaxation of the mean absolute velocity \langle v\rangle (blue) after excitation with a 4\,\mathrm{\tau_{\text{d}}} square pulse for different parameters (A: \Gamma=1.4\,\mathrm{\mu m},L=12\,\mathrm{\mu m} , B: \Gamma=1.4\,\mathrm{\mu m},L=30\,\mathrm{\mu m} , C: \Gamma=0\,\mathrm{\mu m},L=12\,\mathrm{\mu m} , D: \Gamma=0\,\mathrm{\mu m},L=30\,\mathrm{\mu m} ). The black and red curves are the running variance \sigma^{2}_{10}[\langle v\rangle] with a window length of 10 \tau_{\text{d}} and an exponential fit to \sigma^{2}_{10}[\langle v\rangle] , respectively. e Color map of the measured relaxation times \tau_{r} for different combinations of \Gamma and L . Orange symbols indicate the parameter combinations illustrated in A-D. See Fig. S1 for corresponding simulation results. As an experimental realization of a reservoir computing substrate, we employ periodic arrays of hydrodynamically coupled active colloidal oscillators (ACOs) [ 17 ] . Each ACO is formed by steering a self-propelled colloidal particle toward a fixed target position \mathbf{q}_{i} using a feedback-controlled scanning laser. The propulsion velocity \mathbf{v} , and thus the oscillator dynamics, can be continuously tuned by adjusting the laser intensity and steering. A finite feedback delay \tau_{\text{d}} between particle detection and laser repositioning introduces a controlled misalignment in the propulsion direction, generating stable orbital motion around the target (Figs. 1 a-c and Supplementary Movie 1). This setup allows simultaneous control of several hundred oscillators (see Methods). Arranging the oscillators on a hexagonal lattice couples their motion via the flow fields within the liquid and generates complex collective dynamics [ 17 , 18 , 19 ] (Fig. 1 d). In this configuration, the ACOs serve as the nodes of a physical reservoir, where hydrodynamic interactions mediate information processing and shape the network’s nonlinear dynamics [ 8 ] . The coupling strength, and thus the reservoir’s functionality, can be continuously tuned by adjusting the lattice spacing

Read the complete text