Australian Centre for Robotic Vision

Chronic Obstructive Pulmonary Disease (COPD) poses significant public health and economic challenges and performant prognostic models may be useful to direct treatment. The purpose of this study was to develop predictive models, validate the DOSE and updated ADO predictive models, and identify predictors of future hospitalisation and mortality in contemporary Australian COPD patients. Data from 8,578 inpatients and outpatients diagnosed with COPD (via post-bronchodilator spirometry) between 2006 and 2021 at a large South Australian tertiary public hospital were analysed. Multivariate logistic models and Cox regression, utilising penalised regularisation in multiply imputed data, were used to investigate predictors of hospitalisation due to COPD exacerbation at 1-, 3-, and 5-years post-diagnosis, and COPD-specific mortality at 3- and 5-years. Haemoglobin-corrected DLCO (DLCOc) was used to extend the DOSE and updated ADO models. Locally developed models could predict COPD-specific 1-year hospitalisation risk with AUCs 0.80 (95% CI = [0.76, 0.83]) in males and 0.82 (95% CI = [0.78, 0.86]) in females on a temporally distinct hold-out set, with 3- and 5-year AUCs falling within this range. COPD-specific mortality was predicted with AUCs of 0.90 (95% CI = [0.85, 0.94]) and 0.89 (95% CI = [0.84, 0.92]) at 3 and 5 years in females, and 0.90 (95% CI = [0.86, 0.93]) and 0.88 (95% CI = [0.84, 0.92]) in males. Cox regression models predicted survival well in the test set for both females (C-index = 0.88, 95% CI = [0.85, 0.90]) and males (C-index = 0.86, 95% CI = [0.82, 0.88]). Local model performance was superior to that of the DOSE and updated ADO models for all outcomes, although not always significantly. Among the selected predictors, reduced DLCOc was strongly predictive of all outcomes. and acts as a short-term survival risk for follow-up durations less than 10 years. There was no significant difference in performance between sex, and there were differences in selected features and feature strength between sexes. Extending the extant clinical models with DLCOc significantly improved updated ADO and DOSE model fit and improved discriminatory performance, with the extended ADO index achieving AUC of 0.77 (95% CI = [0.75, 0.79]) and 0.87 (95% CI = [0.84, 0.89]) for predicting 5-year hospitalisation and mortality respectively across the full cohort. The extended DOSE index performed similarly with AUCs 0.77 (95% CI = [0.75, 0.79]) and 0.87 (95% CI = 0.83, 0.89)) for 5-year hospitalisation and mortality. Ours is the only large clinical cohort and prognostic study of Australian COPD patients to date. Locally developed models achieved greater discriminative performance than the original updated ADO and DOSE models within our cohort. Extending the ADO and DOSE models with DLCOc significantly improved model fit in our cohort. We recommend further research into the use of DLCOc as a prognostic index for COPD, and its inclusion in future modelling attempts. Retrospectively registered. Clinical trial number: Not Applicable.

Working memory (WM) decline in aging may be related to increases in "neural noise", potentially reflected in the EEG aperiodic exponent. We reanalyzed previously published data to investigate age-related differences in the aperiodic exponent during verbal WM and its relationship with neural inefficiency. EEG was recorded from 24 younger (18-35 years) and 30 older adults (50-86 years) during a modified Sternberg task with 1-letter, 3-letter, and 5-letter load conditions. Younger adults consistently demonstrated steeper aperiodic slopes than older adults. Unexpectedly, both age groups showed decreased (i.e., flattened) aperiodic exponents during retention relative to fixation, with minimal load-dependent effects. Notably, the relationship between task-related exponent changes and WM performance was complex and dependent on the exponent at fixation, particularly in older adults. Finally, flatter exponents during fixation and late retention were associated with greater neural inefficiency during stimulus processing, reflected by increased P3b amplitudes without corresponding WM performance improvements. These findings suggest that flatter exponents are associated with less efficient neural processing and that older adults flexibly modulate their aperiodic exponent during retention to support WM performance.

Several data compressors have been proposed in distributed optimization frameworks of network systems to reduce communication overhead in large-scale applications. In this paper, we demonstrate that effective information compression may occur over time or space during sequences of node communications in distributed algorithms, leading to the concept of spatio-temporal compressors. This abstraction classifies existing compressors and inspires new compressors as spatio-temporal compressors, with their effectiveness described by constructive stability criteria from nonlinear system theory. Subsequently, we incorporate these spatio-temporal compressors <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">directly</i> into standard continuous-time consensus flows and distributed primal-dual flows, establishing conditions ensuring exponential convergence. Additionally, we introduce a novel observer-based distributed primal-dual continuous flow integrated with spatio-temporal compressors, which provides <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">broader</i> convergence conditions. These continuous flows achieve exponential convergence to the global optimum when the objective function is strongly convex and can be discretized using Euler approximations. Finally, numerical simulations illustrate the versatility of the proposed spatio-temporal compressors and verify the convergence of algorithms.

Jitter degrades a spacecraft's fine-pointing ability required for optical communication, earth observation, and space domain awareness. The development of jitter estimation and compensation algorithms requires high-fidelity sensor observations representative of on-board jitter. In this work, we present the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Event-based Star Tracking Under Jitter</i> (e-STURT) dataset-the first event camera-based dataset of star observations under controlled jitter conditions. The special hardware employed for the dataset emulates an event camera undergoing on-board jitter. While the event camera provides asynchronous, high temporal resolution star observations, systematic and repeatable jitter is introduced using a micrometeraccurate piezoelectric actuator. Various jitter sources are simulated using distinct frequency bands and using both axes of motion. Ground-truth jitter is captured in the hardware of the piezoelectric actuator. The resulting dataset consists of 200 sequences and is made publicly available1. This work highlights the dataset generation process, technical challenges, and the resulting limitations. To serve as a baseline, we propose a high-frequency jitter estimation algorithm that operates directly on the event stream. The e-STURT dataset will enable the development of jitter-aware algorithms for mission-critical event-based space-sensing applications.

Artificial intelligence (AI) has vast potential for improving healthcare delivery, but concerns regarding biases in these systems have raised important questions regarding fairness when deployed clinically. Most prior studies on fairness in clinical AI focus solely on performance disparities between subpopulations, which often fall short of connecting the technical outputs of AI systems with sociotechnical outcomes. In this work, we present a simulation-based approach to explore how statistical definitions of algorithmic fairness translate to fairness in long-term outcomes, using AI-assisted breast cancer screening as a case example. We evaluate four fairness criteria and their impact on mortality rates and socioeconomic disparities, while also considering how clinical decision makers' reliance on AI and patients' access to healthcare affect outcomes. Our results highlight how algorithmic fairness does not directly translate into fair and equitable outcomes, underscoring the importance of integrating sociotechnical perspectives to gain a holistic understanding of fairness in healthcare AI.

We study the invertibility of nonlinear dynamical systems from the perspective of contraction and incremental stability analysis and propose a new invertible recurrent neural model: the BiLipREN. In particular, we consider a nonlinear state space model to be robustly invertible if an inverse exists with a state space realisation, and both the forward model and its inverse are contracting, i.e. incrementally exponentially stable, and Lipschitz, i.e. have bounded incremental gain. This property of bi-Lipschitzness implies both robustness in the sense of sensitivity to input perturbations, as well as robust distinguishability of different inputs from their corresponding outputs, i.e. the inverse model robustly reconstructs the input sequence despite small perturbations to the initial conditions and measured output. Building on this foundation, we propose a parameterization of neural dynamic models: bi-Lipschitz recurrent equilibrium networks (biLipREN), which are robustly invertible by construction. Moreover, biLipRENs can be composed with orthogonal linear systems to construct more general bi-Lipschitz dynamic models, e.g., a nonlinear analogue of minimum-phase/all-pass (inner/outer) factorization. We illustrate the utility of our proposed approach with numerical examples.

In open-pit mining, holes are drilled into the surface of the excavation site and are then detonated with explosives to facilitate digging. These blast holes need to be inspected internally for quality assurance, as well as for operational and geological reasons. Manual hole inspection is slow and expensive, limited in its ability to capture the geometric and geological characteristics of holes. This is the motivation behind the development of our autonomous mine site inspection robot—“DIPPeR”. In this paper, the automation aspect of the project is explained. We present a robust navigation and perception framework that provides streamlined blasthole detection, tracking, and precise down-hole sensor insertion during repetitive inspection tasks. To mitigate the effects of noisy GPS and odometry data typical of surface mining environments, we employ a proximity-based adaptive navigation system that enables the vehicle to dynamically adjust its operations according to target detectability and localisation accuracy. For perception, we process LiDAR data to extract the cone-shaped volume of drill waste above ground, and then project the 3D cone points into a virtual depth image to form accurate 2D segmentation of hole regions. To ensure continuous target tracking as the robot approaches the goal, our system automatically adjusts the projection parameters to ensure consistent appearance of the hole in the image. In the vicinity of the hole, we apply least squares circle fitting combined with non-maximum candidate suppression to achieve accurate hole localisation and collision-free down-hole sensor insertion. We demonstrate the effectiveness and robustness of our framework through dedicated perception and navigation feature tests, as well as streamlined mission trials conducted in high-fidelity simulations and real mine-site field experiments.

This study presents a theoretical framework for soil moisture mapping in large agricultural fields by integrating radio signal transmission, Gaussian Process Regression (GPR), and moisture probe data. The proposed approach leverages the strengths of antennas, which offer continuous, non-intrusive coverage across large areas, and moisture probes, which provide ground-truth data. The GPR method is employed to interpolate missing data due to packet loss and reduce noise for antennas and moisture probes. The methodology involves projecting Received Signal Strength Indicator (RSSI) values to estimate average soil moisture along signal paths and dividing the field into blocks to model moisture distribution. A cost-function-based algorithm is developed to balance approximation and estimation errors through weighted parameters, enabling efficient moisture distribution estimation. The study focuses on theoretical validation through simulations and field trials, demonstrating the accuracy and robustness of the approach. The results demonstrate the potential for precision agriculture, offering scalable and cost-effective solutions for monitoring soil moisture.

In this letter, we first present a bias-eliminated weighted (Bias-Eli-W) perspective-n-point (PnP) estimator for stereo visual odometry (VO) with provable consistency. Specifically, we develop a <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\sqrt{n}$</tex-math></inline-formula>-consistent PnP estimator that accounts for 3D point uncertainties, ensuring that the relative pose estimate converges to the true value as the feature number increases. Next, on the stereo VO pipeline side, we propose a framework that triangulates current features for tracking new frames, decoupling temporal dependencies between pose and 3D point errors. We integrate the Bias-Eli-W PnP estimator into the proposed pipeline, creating a synergistic effect that enhances the accuracy of pose estimation. We validate the performance of our method on the KITTI and Oxford RobotCar datasets. Experimental results demonstrate that our method: 1) achieves significant improvements in both relative pose error and absolute trajectory error in large-scale scenarios; 2) provides reliable localization under erratic motions. The successful implementation of the Bias-Eli-W PnP in stereo VO indicates the importance of information screening in robotic estimation tasks, shedding light on diverse applications where PnP is a key ingredient. Our code is open-sourced at <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/LIAS-CUHKSZ/CurrentFeature-Odometry</uri>.

Mobile manipulator robots operating in complex domestic and industrial environments must effectively coordinate their base and arm motions while avoiding obstacles. While current reactive control methods gracefully achieve this coordination, they rely on simplified and idealised geometric representations of the environment to avoid collisions. This limits their performance in cluttered environments. To address this problem, we introduce RMMI, a reactive control framework that leverages the ability of neural Signed Distance Fields (SDFs) to provide a continuous and differentiable representation of the environment’s geometry. RMMI formulates a quadratic program that optimises jointly for robot base and arm motion, maximises the manipulability, and avoids collisions through a set of inequality constraints. These constraints are constructed by querying the SDF for the distance and direction to the closest obstacle for a large number of sampling points on the robot. We evaluate RMMI both in simulation and in a set of real-world experiments. For reaching in cluttered environments, we observe a 25% increase in success rate. For additional details, code, and experiment videos, please visit https://rmmi.github.io/.