Ensuring the stability and safety of humanoid robots during dynamic tasks remains a crucial challenge. The ability to predict and prevent falls is pivotal in such contexts, as falls can hinder performance and waste resources. Existing fall detection and prediction methods for humanoid robots vary in their approaches, and data-driven approaches based on neural networks have shown promising results. Nevertheless, these approaches still lack the efficiency required to provide fall predictions within an acceptable timeframe without compromising accuracy. In this paper, we propose a data-driven approach leveraging a lightweight neural network architecture, enabling accurate fall predictions in near real-time. Our framework, relying on raw IMU sensor data, undergoes comprehensive evaluation across diverse fall scenarios and evaluation metrics. Comparative analyses against baseline architectures from the literature affirm its superior performance in humanoid fall detection and prediction.

Software frameworks for behaviour are critical in robotics as they enable the correct and efficient execution of functions. While modern behaviour systems have improved their composability, they do not focus on smooth transitions and often lack functionality. In this work, we present the Director, a novel behaviour framework that addresses these problems. It has functionality for soft transitions, multiple implementations of the same action chosen based on conditionals, and strict resource control. The system was successfully used in the 2022/2023 Virtual Season and RoboCup 2023 Bordeaux, in the Humanoid Kid Size League. It is implemented at https://github.com/NUbots/DirectorSoccer, which also contains over thirty automated tests and technical documentation on its implementation in NUClear.

We propose a novel optimisation algorithm designed for multi-condition multi-objective problems, utilising a single-step Twin Delayed Deep Deterministic Policy Gradient (TD3) framework with parallel function evaluations. The proposed algorithm's performance was tested on two mathematical test problems and an airfoil shape optimisation scenario. In all cases, our approach outperformed both a previously proposed deep reinforcement learning algorithm and the wellknown NSGA-II genetic algorithm by discovering a more comprehensive Pareto front at a lower number of function evaluations. Additionally, the proposed algorithm demonstrated more stable training of the deep neural networks across all problems.

Rail freight planning problems pose specific challenges that have attracted the attention of academics and industry professionals for many decades. They involve multiple types of assets (trains, stations, terminals, etc.) and are subjected to structural, operational and safety constraints. Even though various approaches have been proposed, few can address the complexity and size of real-world scenarios, and decentralized techniques, like multi-agent systems (MAS), have become more prevalent. The current state of the art in disciplines such as agent technology, reinforcement learning and discrete-event simulation allows the implementation of complex architectures, with multiple actors interacting and learning simultaneously. Therefore, this study takes advantage of these current advances and proposes an innovative approach to real-time traffic management problems in freight railway networks through multi-agent deep reinforcement learning (MADRL). This study was motivated by the decision-making scheduling problems arising in the Hunter Valley Coal Chain (HVCC), located in New South Wales, Australia. The MADRL algorithm uses as the training environment the simulation model currently utilized for capacity planning of the HVCC, allowing experiments with actual data. Thus, we enhanced the simulation model to accommodate a MAS with intelligent agents representing system elements, such as trains, dump stations, and load points. Furthermore, these agents act in a decentralized fashion based on local observations, constituting a partially-observed Markov decision process (dec-POMDP). Three variations of the MADRL approach are presented: a baseline model, an extended model, and one that directly addresses deadlocks. Finally, we present a transfer learning method that improves deadlock resolution and leverages performance. In the experiments, we explore specific, complex scenarios arising in the HVCC, where trains frequently face deadlock conditions. The baseline model outperforms a first-come-first-serve (FCFS) based heuristic used by HVCC's simulation model and a genetic algorithm in instances with up to 60 trains - but fails in more complex scenarios. On the other hand, the most advanced model, which addresses deadlocks via transfer learning, always finds feasible solutions and produces policies that outperform the FCFS-based heuristic in 94% of the instances.

Four-dimensional image-type data can quickly become prohibitively large, and it may not be feasible to directly apply methods, such as persistent homology or convolutional neural networks, to determine the topological characteristics of these data because they can encounter complexity issues. This study aims to determine the Betti numbers of large four-dimensional image-type data. The experiments use synthetic data, and demonstrate that it is possible to circumvent these issues by applying downscaling methods to the data prior to training a convolutional neural network, even when persistent homology software indicates that downscaling can significantly alter the homology of the training data. When provided with downscaled test data, the neural network can estimate the Betti numbers of the original samples with reasonable accuracy.

Emergency department (ED) presentations for mental health (MH) help-seeking have been rising rapidly in recent years. This research aims to identify the service usage demographic for people seeking MH care in the ED, specifically in this case, to understand the usage by First Nation people. This retrospective cohort study examined the sociodemographic and presentation characteristics of individuals seeking MH care in two EDs between 2016 and 2021. Data were collected using existing records and analysed using descriptive univariate analysis with statistical significance between the two sites determined using chi-squared test, p < 0.05. The overall data presented in this analysis show an overall ED mental health presentation rate of 12.02% for those who identified as 'Aboriginal but not Torres Strait Islander origin', 0.36% as 'Both Aboriginal and Torres Strait Islander' and 0.27% as 'Torres Strait Islander' totalling 12.63%. This is an overrepresentation compared to the regional population of 4.9%. One site recorded 14.1% of ED presentations that identified as Aboriginal and/or Torres Strait Islander, over double the site's demographic of 6.3%. Given the disproportionately high representation of First Nation people in MH-related ED presentations, further research is required to prioritise a First Nation research perspective that draws on First Nation research methods, such as yarning and storytelling to understand the unique cultural needs and challenges experienced by First Nation people accessing MH care via ED. Understanding the demographic is but one step in supporting the Cultural Safety needs of First Nation people. Additionally, research should be designed, governed and led by First Nation researchers.

Background: Trustworthiness in Artificial Intelligence (AI) innovation is a priority for governments, researchers and clinicians; however, clinicians have highlighted trust and confidence as barriers to their acceptance of AI within a clinical application. While there is a call to design and develop AI that is considered trustworthy, AI still lacks the emotional capability to facilitate the reciprocal nature of trust.Aim: This paper aims to highlight and discuss the enigma of seeking or expecting trust attributes from a machine and, secondly, reframe the interpretation of trustworthiness for AI through evaluating its reliability and validity as consistent with the use of other clinical instruments.Results: AI interventions should be described in terms of competence, reliability and validity as expected of other clinical tools where quality and safety are a priority. Nurses should be presented with treatment recommendations that describe the validity and confidence of prediction with the final decision for care made by nurses. Future research should be framed to better understand how AI is used to deliver care. Finally, there is a responsibility for developers and researchers to influence the conversation about AI and its power towards improving outcomes.Conclusion: The sole focus on demonstrating trust rather than the business-as-usual requirement for reliability and validity attributes during implementation phases may result in negative experiences for nurses and clinical users.Implications for practice: This research will have significant implications for the way in which future nursing is practised. As AI-based systems become a part of routine practice, nurses will be faced with an increasing number of interventions that require complex trust systems to operate. For any AI researchers and developers, understanding the complexity of trust and creditability in the use of AI in nursing will be crucial for successful implementation. This research will contribute and assist in understanding nurses' role in this change.

Objective The objective of this study was to assess the predictability of admissions to a MH inpatient ward using ML models, based on routine data collected during triage in EDs. This research sought to identify the most effective ML model for this purpose while considering the practical implications of model interpretability for clinical use. Methods The study utilised existing data from January 2016 to December 2021. After data pre-processing, an exploratory analysis revealed the non-linear nature of the dataset. Six different ML models were tested: Random Forest, XGBoost, CatBoost, k-Nearest Neighbours (kNN), Explainable Boosting Machine (EBM) using InterpretML, and Support Vector Machine using Support Vector Classification (SVC). The performance of these models was evaluated using various metrics including the Matthews Correlation Coefficient (MCC). Results Among the models evaluated, the CatBoost model achieved the highest MCC score of 0.1952, demonstrating superior balanced accuracy and predictive power, particularly in correctly identifying positive cases. The InterpretML model also performed well, with an MCC score of 0.1914. While CatBoost showed strong predictive capabilities, its complexity poses challenges for clinical interpretation. Conversely, the InterpretML model, though slightly less powerful, offers better transparency and is more practical for clinical use. Conclusion The findings suggest that the CatBoost model is a compelling choice for scenarios prioritising the detection of positive cases. However, the InterpretML model's ease of interpretation makes it more suitable for clinical application. Integrating explanation methods like SHAP with non-linear models could enhance model transparency and foster clinician trust. Further research is recommended to refine non-linear models within decision support systems, explore multi-source data integration, understand clinician attitudes towards ML, and develop real-time data collection systems. This study highlights the potential of ML in predicting MH admissions from ED data while stressing the importance of interpretability, ethical considerations, and ongoing validation for successful clinical implementation.

This study investigates the potential of Deep Reinforcement Learning (DRL) in the context of Multi-Condition Multi-Objective airfoil shape optimisation by benchmarking a customised DRL algorithm, namely Single-Step Proximal Policy Optimisation, against NSGA-II, a conventional genetic algorithm. We illustrate the capability of the DRL algorithm to effectively optimise across a continuous multi-condition plane, eliminating the need to discretise it into discrete points, a practice commonly employed in conventional Genetic Algorithms. We further demonstrate that the DRL algorithm achieves hypervolume averages and convergence rates that are competitive when compared to NSGA-II. Analysis of Deep Neural Networks extracted from the training phase of the DRL algorithm indicates that almost complete knowledge of the Pareto front is retained by the network, which can be utilised to accelerate the discovery of the Pareto front in similar optimisation tasks via transfer learning.

This paper proposes a low-complexity data-driven multi-user detector for grant-free non-orthogonal multiple access (GF-NOMA), which has gained significant interest in Internet of Things (IoT). IoT traffic is predominantly sporadic, where devices become active whenever they have data to transmit. The conventional grant-access procedure for requesting a transmission slot every time results in significant signaling overhead and latency. In power domain GF-NOMA, multiple devices can be preallocated the same channel resource, but different power levels. Whenever a device has data, it starts transmission directly using the allocated power level without any grant request. While this significantly reduces the signaling overhead, the access point has to perform the complex task of identifying the active devices and decoding their data. Conventional receivers for power domain NOMA fail in such GF scenarios and the typical solution is to limit transmissions to be packet-synchronized and add carefully chosen pilots in every packet to facilitate activity detection. However, in fairly static IoT networks with low-complexity devices and small packet sizes, this represents a significant overhead and reduces efficiency. In this work we solve the GF-NOMA detection problem without these constraints, by analyzing the boundaries of the received constellation points in power domain GF-NOMA for all activation combinations at once. A low-complexity decision tree-based receiver is proposed, which performs as well as the maximum likelihood-based benchmark receiver, and better than traditional data-driven detectors for GF-NOMA. Comprehensive simulation results demonstrate the performance of the proposed detector in terms of its detection efficiency and parameter learning with minimal training data.

This research addresses the critical issue of identifying factors contributing to admissions to acute mental health (MH) wards for individuals presenting to the emergency department (ED) with MH concerns as their primary issue, notably suicidality. This study aims to leverage machine learning (ML) models to assess the likelihood of admission to acute MH wards for this vulnerable population. Data collection for this study used existing ED data from 1 January 2016 to 31 December 2021. Data selection was based on specific criteria related to the presenting problem. Analysis was conducted using Python and the Interpretable Machine Learning (InterpretML) machine learning library. InterpretML calculates overall importance based on the mean absolute score, which was used to measure the impact of each feature on admission. A person's 'Age' and 'Triage category' are ranked significantly higher than 'Facility identifier', 'Presenting problem' and 'Active Client'. The contribution of other presentation features on admission shows a minimal effect. Aligning the models closely with service delivery will help services understand their service users and provide insight into financial and clinical variations. Suicidal ideation negatively correlates to admission yet represents the largest number of presentations. The nurse's role at triage is a critical factor in assessing the needs of the presenting individual. The gap that emerges in this context is significant; MH triage requires a complex understanding of MH and presents a significant challenge in the ED. Further research is required to explore the role that ML can provide in assisting clinicians in assessment.

Understanding the topological characteristics of data is important to many areas of research. Recent work has demonstrated that synthetic 4D image-type data can be useful to train 4D convolutional neural network models to see topological features in these data. These models also appear to tolerate the use of image preprocessing techniques where existing topological data analysis techniques such as persistent homology do not. This paper investigates how methods from algebraic topology, combined with image processing techniques such as morphology, can be used to generate topologically sophisticated and diverse-looking 2-, 3-, and 4D image-type data with topological labels in simulation. These approaches are illustrated in 2D and 3D with the aim of providing a roadmap towards achieving this in 4D.

Emergency Department (ED) presentations for Mental Health (MH) help-seeking have been rising rapidly, with EDs as the main entry point for most individuals in Australia. The objective of this retrospective cohort study was to analyse the sociodemographic and presentation features of people who sought mental healthcare in two EDs located in a regional coastal setting in New South Wales (NSW), Australia from 2016 to 2021. This article is a part of a broader research study on the utilisation of machine learning in MH. The objective of this study is to identify the factors that lead to the admission of individuals to an MH inpatient facility when they seek MH care in an ED. Data were collected using existing records and analysed using descriptive univariate analysis with statistical significance between the two sites was determined using Chi squared test, p < 0.05. Two main themes characterise dominant help-seeking dynamics for MH conditions in ED, suicidal ideation, and access and egress pathways. The main findings indicate that suicidal ideation was the most common presenting problem (38.19%). People presenting to ED who 'Did not wait' or 'Left at own risk' accounted for 10.20% of departures from ED. A large number of presentations arrived via the ambulance, accounting for 45.91%. A large proportion of presentations are related to a potentially life-threatening condition (suicidal ideation). The largest proportion of triage code 1 'Resuscitation' was for people with presenting problem of 'Behavioural Disturbance'. Departure and arrival dynamics need to be better understood in consultation with community and lived experience groups to improve future service alignment with the access and egress pathways for emergency MH care.

Zero-shot learning enables models to generalise to unseen classes using semantic information, bridging the gap between training classes and previously unseen test classes. While widely studied in computer vision, its application to environmental audio remains underexplored, and generative approaches have received little attention. This work presents the first benchmark of generative methods for zero-shot environmental sound classification. Four approaches spanning variational, adversarial, diffusion-based, and denoising paradigms are evaluated. The benchmark includes CADA-VAE and LisGAN, adapted from computer vision, together with two embedding-generation methods introduced in this work: one based on a denoising diffusion probabilistic model (DDPM) and the other on a conditional generative denoising network (CGDN). Experiments on five environmental audio datasets (ESC-50, ARCA23K-FSD, FSC22, UrbanSound8K, and TAU Urban Acoustic Scenes 2019) and one music dataset (GTZAN) show that generative methods are competitive with established compatibility-based approaches. Among the evaluated generative methods, CGDN achieves the highest average accuracy and is the only one to significantly outperform both the DDPM- and GAN-based methods, while remaining statistically indistinguishable from the strong ALE baseline. These findings suggest that optimisation stability is an important factor in generative zero-shot learning for environmental audio.

Because of the number of different architectures, numerous settings of their hyper-parameters and disparity among their sizes, it is difficult to equitably compare various deep neural network (DNN) architectures for sentence classification. Evolutionary algorithms are emerging as a popular method for the automatic selection of architectures and hyperparameters for DNNs whose generalisation performance is heavily impacted by such settings. Most of the work in this area is done in the image domain, leaving text analysis, another prominent application domain of deep learning, largely absent. Besides, literature presents conflicting claims regarding the superiority of one DNN architecture over others in the context of sentence classification. To address this issue, we propose a genetic algorithm (GA) for optimising the architectural and hyperparameter settings in different DNN types for sentence classification. To enable the representation of the wide variety of architectures and hyperparameters utilised in DNNs, we employed a generalised and flexible encoding scheme in our GA. Our study involves optimising two convolutional and three recurrent architectures to ensure a fair and unbiased evaluation of their performance. Furthermore, we explore the effects of using F1 score versus accuracy as a performance metric during evolutionary optimisation of those architectures. Our results, using ten datasets, show that, in general, the architectures and hyperparameters evolved using the F1 score tended to outperform those evolved using accuracy and in the case of CNN and BiLSTM the results were significant in statistical measures. Of the five architectures considered, the GA-evolved gated recurrent unit (GRU) performed the strongest overall, achieving good generalisation performance while using relatively few trainable parameters, establishing GRU as the preferred architecture for the sentence classification task. The optimised architectures exhibited comparable performance with the state-of-the-art, given the large difference in trainable parameters.

An integrative review investigating the incorporation of artificial intelligence (AI) and machine learning (ML) based decision support systems in mental health care settings was undertaken of published literature between 2016 and 2021 across six databases. Four studies met the research question and the inclusion criteria. The primary theme identified was trust and confidence. To date, there is limited research regarding the use of AI-based decision support systems in mental health. Our review found that significant barriers exist regarding its incorporation into practice primarily arising from uncertainty related to clinician's trust and confidence, end-user acceptance and system transparency. More research is needed to understand the role of AI in assisting treatment and identifying missed care. Researchers and developers must focus on establishing trust and confidence with clinical staff before true clinical impact can be determined. Finally, further research is required to understand the attitudes and beliefs surrounding the use of AI and related impacts for the wellbeing of the end-users of care. This review highlights the necessity of involving clinicians in all stages of research, development and implementation of artificial intelligence in care delivery. Earning the trust and confidence of clinicians should be foremost in consideration in implementation of any AI-based decision support system. Clinicians should be motivated to actively embrace the opportunity to contribute to the development and implementation of new health technologies and digital tools that assist all health care professionals to identify missed care, before it occurs as a matter of importance for public safety and ethical implementation. AI-basesd decision support tools in mental health settings show most promise as trust and confidence of clinicians is achieved.

One factor that limits development of fundamental research on the influence of coke microstructure on its strength is the difficulty in quantifying the way that microstructure is both classified and distributed in three dimensions. To support such fundamental studies, this study evaluated a novel volumetric approach for classifying small (approx. 450 μm 3 ) blocks of coke microstructure from 3D computed tomography scans. An automated process for classifying microstructure blocks was described. It is based on Landmark Multi-Dimensional Scaling and uses the Bhattacharyya metric and k-means clustering. The approach was evaluated using 27 coke samples across a range of coke with different properties and reliably identified 6 ordered class of coke microstructure based on the distribution of voxel intensities associated with structural density. The lower class (1–2) subblocks tend to be dominated by pores and thin walls. Typically, there is an increase in wall thickness and reduced pore sizes in the higher classes. Inert features are also likely to be seen in higher classes (5–6). In general, this approach provides an efficient automated means for identifying the 3D spatial distribution of microstructure in CT scans of coke.

The automatic detection of violent behavior in acoustic data has become an important research area due to its growing application potential in various surveillance and behavior monitoring tasks where it causes fewer privacy issues than video. In this paper, we propose two deep learning approaches of different characteristics to classify speech data containing violent behavior against data that only includes patterns of non-violent behavior. The first approach is based on conventional deep neural networks, while the second approach uses lightweight deep neural networks. Both utilize Mel-spectrogram images of speech signals as input to fine-tuned models. Our best lightweight model's classification accuracy is about 8% better than the previous state-of-the-art result using the same benchmark data. Lightweight models have fewer parameters and require fewer computing resources than conventional deep models. This can be an important advantage when being deployed on mobile or edge devices.

Background. The prominence of technology in modern life cannot be understated. However, for some people, these innovations or their related plausible advancements can be associated with perceptual misinterpretation and/or incorporation into delusional concepts. Objective. This paper aims to explore the intersection of technological advancement and experiencing psychosis. We present a discussion about the explanation seeking that incorporates the concept, that for some people, of technological innovation becoming intertwined with delusional symptoms over the past 100 years. Methods. A longitudinal review of the literature was conducted to synthesize and draw these concepts together, mapping them to a timeline that aligns computing science and healthcare expertise and presents the significant technological changes of the modern era charted against mental health milestones and reports of technology-related delusions. Results. It is possible for technology to be incorporated into the content of delusions with evidence supporting a link between the rate of technological change, the content of delusions, and the use of technology as a way of seeking an explanation. Moreover, analysis suggests a need to better understand how innovations may impact the mental health of people at risk of psychosis and other mental health conditions. Conclusions. Clinical experts and lived experience experts need to be informed about and collaborate with future research and development of technology, specifically artificial intelligence and machine learning, early in the development cycle. This concurs with other artificial intelligence research recommendations calling for design attention to the development and implementation of technological innovation applied in a mental health context.

Zero-shot learning is a scenario in machine learning where the classes used in the training and test sets are disjoint. This work considers zero-shot learning for environmental audio and improves results by enhancing audio and word embeddings. Previous works use the VGGish model for audio embeddings, and textual class labels are often used as input for word embedding networks such as Word2Vec. This study instead uses a modified YAMNet network to obtain semantic audio embeddings for zero-shot learning. Moreover, part of this study involves adding linguistic devices, such as synonyms, semantic broadening and onomatopoeia, to the input of the word embeddings. With these two modifications, top-1 accuracy is increased on average by over five percentage points compared to the state-of-the-art on ESC-50. This emerging area of research has applications in robot awareness, security systems and wildlife conservation in situations where no data is available for some classes.