Maritime Autonomous Surface Ships (MASS) require assurance methods that can maintain justified confidence as autonomy software, operational contexts, and regulatory expectations evolve. Current verification and certification practices remain largely design-time and document-centred, and therefore struggle to scale to adaptive, data-driven, and software-intensive maritime autonomy. This paper proposes a conceptual framework and research roadmap for applying Agentic AI to the assurance of maritime autonomy. The framework treats assurance as a lifecycle process in which specialised agents collect evidence, maintain structured world and knowledge models, perform bounded assurance inference, coordinate verification activities, and escalate uncertain cases to human assurers. The proposed architecture combines an eleven-layer agentic assurance loop with a three-tier multi-agent structure spanning component-level, coordinator-level, and system-level assurance agents. A bounded Autonomous Navigation System (ANS) verification and validation use case illustrates how such agents can organise subsystem evidence, compute traceable assurance indicators, separate deterministic PASS/FAIL/REVIEW decisions from LLM-based explanations, and support human review. The prototype is not intended as industrial validation or as evidence of regulatory compliance; but to demonstrate how the proposed concepts can be operationalised in a lightweight assurance workflow. The paper concludes with a staged roadmap covering formalisation of assurance contracts and evidence objects, calibrated confidence models, digital-twin testing, regulatory traceability, human-assurer evaluation, and industrial validation.
Transitional foods (TFs) are foods that change texture with moisture and/or temperature. Despite their potential to aid chewing development and swallowing rehabilitation, TFs remain underutilized in clinical practice. This study explored the perspectives of speech and language therapists (SLTs) in Singapore regarding TFs in dysphagia and feeding management, aiming to inform the development of TF products to expand available food options. A cross-sectional online survey was conducted among actively practicing SLTs. The survey included closed and open-ended questions on understanding of TFs and its usage, and perceived facilitators and barriers to the use of TFs. Fifty-eight registered SLTs who had managed clients with feeding difficulties and/or dysphagia within the last five years participated. Most worked in public health institutions with adult clients. While 93.1% had heard of TFs, only 22.4% accurately described them per the International Dysphagia Diet Standardisation Initiative (IDDSI) definition; 56.9% provided partially correct descriptions lacking key defining features, and 12.1% gave erroneous definitions. Over half (58.6%) had recommended TFs, primarily to expand food options and support therapy, despite infrequent client-driven requests for TFs. Key barriers included concerns about swallowing safety, variability in texture transitions, limited training, and institutional constraints. The findings highlight a gap in SLTs' knowledge and confidence regarding TFs. Despite the perceived benefits of TFs, unclear definitions, lack of safety data, and systemic barriers hinder consistent use. The broad IDDSI definition may limit clinical applicability; refining it through subcategorization based on specific characteristics could improve safety assessment and support wider adoption in dysphagia and feeding management.
Large Audio-Language Models (LALMs) enable general audio understanding and demonstrate remarkable performance across various audio tasks. However, these models still face challenges in temporal perception (e.g., inferring event onset and offset), leading to limited utility in fine-grained scenarios. To address this issue, we propose Audio-Side Time Prompt and leverage Reinforcement Learning (RL) to develop the TimePro-RL framework for fine-grained temporal perception. Specifically, we encode timestamps as embeddings and interleave them within the audio feature sequence as temporal coordinates to prompt the model. Furthermore, we introduce RL following Supervised Fine-Tuning (SFT) to directly optimize temporal alignment performance. Experiments demonstrate that TimePro-RL achieves significant performance gains across a range of audio temporal tasks, such as audio grounding, sound event detection, and dense audio captioning, validating its robust effectiveness.
Spectrogram-based representations have grown to dominate the feature space for deep learning audio analysis systems, and are often adopted for speech analysis also. Initially, the primary motivator for spectrogram-based representations was their ability to present sound as a two dimensional signal in the time-frequency plane, which not only provides an interpretable physical basis for analysing sound, but also unlocks the use of a wide range of machine learning techniques such as convolutional neural networks, that had been developed for image processing. A spectrogram is a matrix characterised by the resolution and span of its two dimensions, as well as by the representation and scaling of each element. Many possibilities for these three characteristics have been explored by researchers across numerous application areas, with different settings showing affinity for various tasks. This paper reviews the use of spectrogram-based representations and surveys the state-of-the-art to question how front-end feature representation choice allies with back-end classifier architecture for different tasks.
Robotic compliance control is critical for delicate tasks such as electronic connector assembly, where precise force regulation and adaptability are paramount. However, traditional methods often struggle with modeling inaccuracies and sensor noise. Inspired by human adaptability in complex assembly operations, we present RoboMT, a novel framework that integrates a Mamba algorithm with a Transformer architecture to achieve human-like compliance control. By leveraging a bilateral teleoperation platform, we collect extensive real-time force/torque and motion data to form a comprehensive dataset for training. Furthermore, RoboMT incorporates an Adaptive Action Chunk module and a Temporal Fusion module to ensure smooth and robust action prediction. Experimental results across four electronic assembly tasks show that RoboMT achieves superior success rates (6298%) over baselines (2998%), while maintaining stable force regulation around 2.5N, closely resembling human performance. During task transitions, RoboMT quickly stabilizes at 5N with minimal overshoot, avoiding the large force spikes (over 24N) seen in baselines. Additionally, RoboMT maintains an average inference speed of 55 ms per batch, balancing real-time responsiveness and control robustness. Overall, RoboMT presents a compelling pathway toward error-minimized, human-level compliance control, and generalization for real-world robotic assembly, setting a new benchmark for precision, adaptability, and robustness in robotic assembly.
