arXiv subjects

Michael Groom

Publications and source records attributed to Michael Groom.

15 recordsLinked to original sources

arXiv2026

Maximising the Set-Piece Return: Optimising Football Corner Tactics with Graph Reinforcement Learning

Machine learning is increasingly employed for the evaluation of football tactics. However, existing approaches focus on characterising historical actions or analyst-specified counterfactual scenarios. In this work, we seek to go beyond the imitation of historically observed patterns towards discovering new generalisable player configurations and strategies. To tackle this, we focus on optimising corner kick routines, and formulate a decision-making problem in which a central policy makes adjustments to attacking player positions and velocities to maximise first contact shot probability. Unlike classic optimisation that solves for isolated setups, we contribute a reinforcement learning architecture operating on graph-structured data that yields a general policy for adjusting arbitrary starting player positions. Evaluated on over 3,000 Premier League corners, our approach strongly outperforms baseline optimisation techniques under matched inference budgets. Our results suggest that graph reinforcement learning can shift set-piece analysis from historical evaluation and imitation towards reward-driven tactical discovery.

cs.LG↗

arXiv2026

Quantile-Coupled Flow Matching for Distributional Reinforcement Learning

Unlike standard expected-return Reinforcement Learning (RL), Distributional RL (DRL) models the full return distribution, making it better-suited for uncertainty-aware and risk-sensitive decision-making. Conditional Flow Matching (CFM) critics have recently attracted attention for modelling continuous, multi-modal return distributions. Despite this interest, there remains a substantial metric mismatch: DRL theory relies on the distributional Bellman operator being contractive in the $p$-Wasserstein distance, yet existing CFM critics are trained with arbitrary source-target couplings, so their flow-matching losses are not Wasserstein-aligned surrogates for matching Bellman target return distributions. In this work, we address this mismatch by proposing FlowIQN, a CFM critic that sorts source and Bellman target samples within each mini-batch to approximate the monotone optimal transport coupling, replacing arbitrary pairings with quantile-aligned flow paths. We prove that the loss of our quantile-coupled CFM critic yields a Wasserstein-aligned approximate projection compatible with the foundations of DRL. To our knowledge, FlowIQN is the first flow-matching distributional critic with an explicit Wasserstein-aligned projection guarantee. We further extend FlowIQN with shortcut models for efficient inference. Empirical results show that FlowIQN improves Wasserstein return-distribution accuracy over other CFM critics. It also yields competitive performance on offline RL benchmarks across multiple policy extraction methods, providing a theoretically grounded CFM critic that is readily compatible with DRL pipelines. Code: https://github.com/ori-goals/flowIQN.

cs.LG↗

arXiv2026

Risk-Aware Reinforcement Learning for Mobile Manipulation

For robots to successfully transition from lab settings to everyday environments, they must begin to reason about the risks associated with their actions and make informed, risk-aware decisions. This is particularly true for robots performing mobile manipulation tasks, which involve both interacting with and navigating within dynamic, unstructured spaces. However, existing whole-body controllers for mobile manipulators typically lack explicit mechanisms for risk-sensitive decision-making under uncertainty. To our knowledge, we are the first to (i) learn risk-aware visuomotor policies for mobile manipulation conditioned on egocentric depth observations with runtime-adjustable risk sensitivity, and (ii) show risk-aware behaviours can be transferred through Imitation Learning (IL) to a visuomotor policy conditioned on egocentric depth observations. Our method achieves this by first training a privileged teacher policy using Distributional Reinforcement Learning (DRL), with a risk-neutral distributional critic. Distortion risk-metrics are then applied to the critic's predicted return distribution to calculate risk-adjusted advantage estimates used in policy updates to achieve a range of risk-aware behaviours. We then distil teacher policies with IL to obtain risk-aware student policies conditioned on egocentric depth observations. We perform extensive evaluations demonstrating that our trained visuomotor policies exhibit risk-aware behaviour (specifically achieving better worst-case performance) while performing reactive whole-body motions in unmapped environments, leveraging live depth observations for perception.

cs.RO↗

arXiv2026

Distillation and Interpretability of Ensemble Forecasts of ENSO Phase using Entropic Learning

This paper introduces a distillation framework for an ensemble of entropy-optimal Sparse Probabilistic Approximation (eSPA) models, trained exclusively on satellite-era observational and reanalysis data to predict ENSO phase up to 24 months in advance. While eSPA ensembles yield state-of-the-art forecast skill, they are harder to interpret than individual eSPA models. We show how to compress the ensemble into a compact set of "distilled" models by aggregating the structure of only those ensemble members that make correct predictions. This process yields a single, diagnostically tractable model for each forecast lead time that preserves forecast performance while also enabling diagnostics that are impractical to implement on the full ensemble. An analysis of the regime persistence of the distilled model "superclusters", as well as cross-lead clustering consistency, shows that the discretised system accurately captures the spatiotemporal dynamics of ENSO. By considering the effective dimension of the feature importance vectors, the complexity of the input space required for correct ENSO phase prediction is shown to peak when forecasts must cross the boreal spring predictability barrier. Spatial importance maps derived from the feature importance vectors are introduced to identify where predictive information resides in each field and are shown to include known physical precursors at certain lead times. Case studies of key events are also presented, showing how fields reconstructed from distilled model centroids trace the evolution from extratropical and inter-basin precursors to the mature ENSO state. Overall, the distillation framework enables a rigorous investigation of long-range ENSO predictability that complements real-time data-driven operational forecasts.

physics.comp-ph↗

arXiv2026

Multi-Scale Wavelet Transformers for Operator Learning of Dynamical Systems

Recent years have seen a surge in data-driven surrogates for dynamical systems that can be orders of magnitude faster than numerical solvers. However, many machine learning-based models such as neural operators exhibit spectral bias, attenuating high-frequency components that often encode small-scale structure. This limitation is particularly damaging in applications such as weather forecasting, where misrepresented high frequencies can induce long-horizon instability. To address this issue, we propose multi-scale wavelet transformers (MSWTs), which learn system dynamics in a tokenized wavelet domain. The wavelet transform explicitly separates low- and high-frequency content across scales. MSWTs leverage a wavelet-preserving downsampling scheme that retains high-frequency features and employ wavelet-based attention to capture dependencies across scales and frequency bands. Experiments on chaotic dynamical systems show substantial error reductions and improved long horizon spectral fidelity. On the ERA5 climate reanalysis, MSWTs further reduce climatological bias, demonstrating their effectiveness in a real-world forecasting setting.

cs.LG↗

arXiv2025

An entropy-optimal path to humble AI

Progress of AI has led to very successful, but by no means humble models and tools, especially regarding (i) the huge and further exploding costs and resources they demand, and (ii) the over-confidence of these tools with the answers they provide. Here we introduce a novel mathematical framework for a non-equilibrium entropy-optimizing reformulation of Boltzmann machines based on the exact law of total probability and the exact convex polytope representations. We show that it results in the highly-performant, but much cheaper, gradient-descent-free learning framework with mathematically-justified existence and uniqueness criteria, and cheaply-computable confidence/reliability measures for both the model inputs and the outputs. Comparisons to state-of-the-art AI tools in terms of performance, cost and the model descriptor lengths on a broad set of synthetic and real-world problems with varying complexity reveal that the proposed method results in more performant and slim models, with the descriptor lengths being very close to the intrinsic complexity scaling bounds for the underlying problems. Applying this framework to historical climate data results in models with systematically higher prediction skills for the onsets of important La Ni\~na and El Ni\~no climate phenomena, requiring just few years of climate data for training - a small fraction of what is necessary for contemporary climate prediction tools.

cs.LG↗

arXiv2025

Entropic learning enables skilful forecasts of ENSO phase at up to two years lead time

This paper extends previous work (Groom et al., \emph{Artif. Intell. Earth Syst.}, 2024) in applying the entropy-optimal Sparse Probabilistic Approximation (eSPA) algorithm to predict ENSO phase, defined by thresholding the Ni\~no3.4 index. Only satellite-era observational datasets are used for training and validation, while retrospective forecasts from 2012 to 2022 are used to assess out-of-sample skill at lead times up to 24 months. Rather than train a single eSPA model per lead, we introduce an ensemble approach in which multiple eSPA models are aggregated via a novel meta-learning strategy. The features used include the leading principal components from a delay-embedded EOF analysis of global sea surface temperature, vertical temperature gradient (a thermocline proxy), and tropical Pacific wind stresses. Crucially, the data is processed to prevent any form of information leakage from the future, ensuring realistic real-time forecasting conditions. Despite the limited number of training instances, eSPA avoids overfitting and produces probabilistic forecasts with skill comparable to the International Research Institute for Climate and Society (IRI) ENSO prediction plume. Beyond the IRI's lead times, eSPA maintains skill out to 22 months for the ranked probability skill score and 24 months for accuracy and area under the ROC curve, all at a fraction of the computational cost of a fully-coupled dynamical model. Furthermore, eSPA successfully forecasts the 2015/16 and 2018/19 El Ni\~no events at 24 months lead, the 2016/17, 2017/18 and 2020/21 La Ni\~na events at 24 months lead and the 2021/22 and 2022/23 La Ni\~na events at 12 and 8 months lead.

physics.comp-ph↗

arXiv2025

Impact of transverse strain on linear, transitional and self-similar turbulent mixing layers

The growth of interfacial instabilities such as the Rayleigh-Taylor (RTI) and Richtmyer-Meshkov instability (RMI) are modified when developing in convergent geometries. Whilst these modifications are usually quantified by the compression rate and convergence rate of the mixing layer, an alternative framework is proposed, describing the evolution of the mixing layer through the effects of the mean strain rates experienced by the mixing layer. An investigation into the effect of the transverse strain rate on the mixing layer development is conducted through application of transverse strain rates in planar geometry. A model for the linear regime in planar geometry with transverse strain rate is derived, with equivalent solutions to convergent geometry, and validated with two-dimensional simulations demonstrating the amplification of the instability growth under transverse compression. The effect of the transverse strain rate on the transitional-to-turbulent mixing layer is investigated with implicit large eddy simulation based on the multi-mode quarter-scale $\theta$-group case by Thornber et al. (Phys. Fluids, vol. 29, 2017, 105107). The mixing layer's growth exhibits the opposite trend to the linear regime model, with reduced growth under transverse compression. The effect of shear-production under transverse compression causes the mixing layer to become more mixed and the turbulent kinetic energy is increasingly dominated by the transverse directions, deviating from the unstrained self-similar state. The mixing layer width is able to be predicted by adjusting the buoyancy-drag model by Youngs & Thornber (Physica D, vol. 410, 2020, 132517) to utilise a drag length scale that scales with the transverse expansion.

physics.flu-dyn↗

arXiv2025

Turbulence Modelling of Mixing Layers under Anisotropic Strain

The development of turbulent mixing layers can be altered by the application of anisotropic strain rates, potentially arising from radial motion in convergent geometry or movement through non-uniform geometry. Previous closure models and calibrations of compressible turbulence models tend to focus on incompressible flows or isotropic strain cases, which is in contrast to many real flow conditions. The treatment of bulk compression under anisotropic strain is investigated using the K-L turbulence model, a two-equation Reynolds-Averaged Navier-Stokes (RANS) model that is commonly used for simulating interfacial instabilities. One-dimensional simulations of shock-induced turbulent mixing layers under applied axial or transverse strain rates are performed using three different closures for the bulk compression of the turbulent length scale. The default closure method using the mean isotropic strain rate is able to reasonably predict the integral width and turbulent kinetic energy of the mixing layer under the applied strain rates. However, the K-L model's performance is improved with the transverse strain closure, while the axial strain closure worsens the model. The effects of this new closure are investigated for the buoyancy-drag model, showing that a three-equation model which evolves the integral width and turbulent length scale separately is most effective for modelling anisotropic strain. Through the equivalence of two-equation RANS models, the modification of the bulk compression closure for the turbulent length scale also suggests an alteration to the K-$\epsilon$ and K-$\omega$ models.

physics.flu-dyn↗

arXiv2024

COBRA-PPM: A Causal Bayesian Reasoning Architecture Using Probabilistic Programming for Robot Manipulation Under Uncertainty

Manipulation tasks require robots to reason about cause and effect when interacting with objects. Yet, many data-driven approaches lack causal semantics and thus only consider correlations. We introduce COBRA-PPM, a novel causal Bayesian reasoning architecture that combines causal Bayesian networks and probabilistic programming to perform interventional inference for robot manipulation under uncertainty. We demonstrate its capabilities through high-fidelity Gazebo-based experiments on an exemplar block stacking task, where it predicts manipulation outcomes with high accuracy (Pred Acc: 88.6%) and performs greedy next-best action selection with a 94.2% task success rate. We further demonstrate sim2real transfer on a domestic robot, showing effectiveness in handling real-world uncertainty from sensor noise and stochastic actions. Our generalised and extensible framework supports a wide range of manipulation scenarios and lays a foundation for future work at the intersection of robotics and causality.

cs.RO↗

arXiv2023

Numerical Simulation of an Idealised Richtmyer-Meshkov Instability Shock Tube Experiment

The effects of initial conditions on the evolution of the Richtmyer-Meshkov instability (RMI) at early to intermediate times are analysed, using numerical simulations of an idealised version of recent shock tube experiments performed at the University of Arizona. The experimental results are bracketed by performing both implicit large-eddy simulations of the high-Reynolds-number limit as well as direct numerical simulations (DNS) at Reynolds numbers lower than those observed in the experiments. Various measures of the mixing layer width, based on both the plane-averaged turbulent kinetic energy and volume fraction profiles are used to explore the effects of initial conditions on $θ$ and are compared with the experimental results. The decay rate of the total fluctuating kinetic energy is also used to estimate $θ$ based on a relationship that assumes self-similar growth of the mixing layer. The estimates for $θ$ range between 0.44 and 0.52 for each of the broadband perturbations considered and are in good agreement with the experimental results. Overall, the results demonstrate important differences between broadband and narrowband surface perturbations, as well as persistent effects of finite bandwidth on the growth rate of mixing layers evolving from broadband perturbations. Good agreement is obtained with the experiments for the different quantities considered; however, the results also show that care must be taken when using measurements based on the velocity field to infer properties of the concentration field.

physics.flu-dyn↗

arXiv2023

Reynolds number dependence of turbulence induced by the Richtmyer-Meshkov instability using direct numerical simulations

This paper investigates the Reynolds number dependence of a turbulent mixing layer evolving from the Richtmyer-Meshkov instability using a series of direct numerical simulations of a well-defined narrowband initial condition for a range of different Reynolds numbers. The growth rate exponent of the integral width and mixed mass is shown to marginally depend on the initial Reynolds number Re0, as does the minimum value of the molecular mixing fraction. The decay rates of turbulent kinetic energy and its dissipation rate are shown to decrease with increasing Re0, while the spatial distribution of these quantities is biased towards the spike side of the layer. The normalised dissipation rate and scalar dissipation rate are calculated and are observed to be approaching a high Reynolds number limit. By fitting an appropriate functional form, the asymptotic value of these two quantities is estimated as 1.54 and 0.66. Finally, an evaluation of the mixing transition criterion for unsteady flows is performed, showing that even for the highest Re0 case the turbulence in the flow is not yet fully developed. This is despite the observation of a narrow inertial range in the turbulent kinetic energy spectra, with a scaling close to -3/2.

physics.flu-dyn↗

arXiv2020

The influence of initial perturbation power spectra on the growth of a turbulent mixing layer induced by Richtmyer-Meshkov instability

This paper investigates the influence of different broadband perturbations on the evolution of a Richtmyer--Meshkov turbulent mixing layer initiated by a Mach 1.84 shock traversing a perturbed interface separating gases with a density ratio of 3:1. Both the bandwidth of modes in the interface perturbation, as well as their relative amplitudes, are varied in a series of carefully designed numerical simulations at grid resolutions up to $3.2\times10^9$ cells. Three different perturbations are considered, characterised by a power spectrum of the form $P(k)\propto k^m$ where $m=-1$, $-2$ and $-3$. The growth of the mixing layer is shown to strongly depend on the initial conditions, with the growth rate exponent $θ$ found to be $0.5$, $0.63$ and $0.75$ for each value of $m$ at the highest grid resolution. The asymptotic values of the molecular mixing fraction $Θ$ are also shown to vary significantly with $m$; at the latest time considered $Θ$ is $0.56$, $0.39$ and $0.20$ respectively. Turbulent kinetic energy (TKE) is also analysed in both the temporal and spectral domains. The temporal decay rate of TKE is found not to match the predicted value of $n=2-3θ$, which is shown to be due to a time-varying {normalised dissipation rate $C_ε$}. In spectral space, the data follow the theoretical scaling of $k^{(m+2)/2}$ at low wavenumbers and tend towards $k^{-3/2}$ and $k^{-5/3}$ scalings at high wavenumbers for the spectra of transverse and normal velocity components respectively. The results represent a significant extension of previous work on the Richtmyer--Meshkov instability evolving from broadband initial perturbations and provide useful benchmarks for future research.

physics.flu-dyn↗

arXiv2019

Direct numerical simulation of the multimode narrowband Richtmyer-Meshkov instability

Early to intermediate time behaviour of the planar Richtmyer--Meshkov instability (RMI) is investigated through direct numerical simulation (DNS) of the evolution of a deterministic interfacial perturbation initiated by a $\textrm{Ma}=1.84$ shock. The model problem is the well studied initial condition from the recent $θ$-group collaboration [Phys. Fluids. 29 (2017) 105107]. A grid convergence study demonstrates that the Kolmogorov microscales are resolved by the finest grid for the entire duration of the simulation, and that both integral and spectral quantities of interest are converged. Comparisons are made with implicit large eddy simulation (ILES) results from the $θ$-group collaboration, generated using the same numerical algorithm. The total amount of turbulent kinetic energy (TKE) is shown to be decreased in the DNS compared to the ILES, particularly in the transverse directions, giving rise to a greater level of anisotropy in the flow (70\% vs. 40\% more TKE in the shock parallel direction at the latest time considered). This decrease in transfer of TKE to the transverse components is shown to be due to the viscous suppression of secondary instabilities. Overall the agreement in the large scales between the DNS and ILES is very good and hence the mixing width and growth rate exponent $θ$ are very similar. There are substantial differences in the small scale behaviour however, with a 38\% difference observed in the minimum values obtained for the mixing fractions $Θ$ and $Ξ$. Differences in the late time decay of TKE are also observed, with decay rates calculated to be $τ^{-1.41}$ and $τ^{-1.25}$ for the DNS and ILES respectively.

physics.comp-ph↗

arXiv2018

A five-equation model for the simulation of miscible and viscous compressible fluids

Typical multispecies compressible Navier-Stokes computations employ conservative equations for mass fraction transport. Upwind discretisations of these governing equations produce spurious pressure oscillations at diffuse contact surfaces between gases of differing ratio of specific heat capacities which degrade the convergence rate of the algorithm. Adding quasi-conservative equations for volume fraction can solve this error, however this approach has been derived only for immiscible fluids. Here, a five-equation quasi-conservative model is proposed that includes the effects of species diffusion, viscosity and thermal conductivity. The derivation of the model is presented, along with a numerical method to solve the governing equations at second order accuracy in space and time. Formal convergence studies demonstrate the expected order of accuracy is achieved for three benchmark problems, cross-validated against two standard mass fraction models. In these test cases, the new model has between 2 and 10 times lower error for a given grid size. Simulations of a two-dimensional air-SF_6 Richtmyer-Meshkov instability demonstrate that the new model converges to the solution with four times fewer points in each direction when compared to the mass fraction model in an identical numerical framework. This represents an ~40 times lower computational cost for an equivalent error in two-dimensional computations. The proposed model is thus very suitable for Direct Numerical Simulation and Large Eddy Simulation of compressible mixing.

physics.comp-ph↗