Abstract

Dual-arm robotic grasping is crucial for handling large objects that require stable and coordinated manipulation. While single-arm grasping has been extensively studied, datasets tailored for dual-arm settings remain scarce. We introduce a large-scale dataset of 16 million dual-arm grasps, evaluated under improved force-closure constraints. Additionally, we develop a benchmark dataset containing 300 objects with approximately 30,000 grasps, evaluated in a physics simulation environment, providing a better grasp quality assessment for dual-arm grasp synthesis methods. Finally, we demonstrate the effectiveness of our dataset by training a Dual-Arm Grasp Classifier network that outperforms the state of-the-art methods by 15%, achieving higher grasp success rates and improved generalization across objects.

Abstract

Lane detection is critical for Advanced Driver Assistance Systems (ADAS) in autonomous vehicles. Deep learning models offer high accuracy but are computationally demanding for resource-constrained edge platforms, while lightweight image processing algorithms are energy-efficient but often falter in complex scenarios. This paper presents RL-ALMS (Reinforcement Learning-based Adaptive and Lightweight Model Selection), a framework that dynamically switches between deep learning and image processing lane detection models in electric vehicles (EVs). RL-ALMS employs a contextual multi-armed bandit with Thompson sampling to intelligently select the optimal model based on road conditions, battery state, and accuracy needs. The novelty lies in its system-level integration of a battery-aware reward function and a sustainability-driven selection mechanism, specifically designed to balance real-time detection accuracy with long-term EV energy constraints. Experimental results on an edge computing platform demonstrate RL-ALMS achieves 90.3% average accuracy over a 50km journey while maintaining 81.2% battery reserve, significantly outperforming fixed high-accuracy (battery depletion) and energy-efficient (64.5% accuracy) strategies. RL-ALMS maintains consistent performance across diverse conditions, representing a practical advancement for energy-constrained, real-time lane detection in EVs.

Abstract

This study addresses the critical challenge of data scarcity in machine translation for Indian languages, particularly given their morpholog- ical complexity and limited parallel data. We investigate an effective strategy to maximize the utility of existing data by generating nega- tive samples from positive training instances us- ing a progressive perturbation approach. This is used to align the model with preferential data using Kahneman-Tversky Optimization (KTO). Comparing it against traditional Su- pervised Fine-Tuning (SFT), we demonstrate how generating negative samples and leverag- ing KTO enhances data efficiency. By creat- ing rejected samples through progressively per- turbed translations from the available dataset, we fine-tune the Llama 3.1 Instruct 8B model using QLoRA across 16 language directions, in- cluding English, Hindi, Bangla, Tamil, Telugu, and Santali. Our results show that KTO-based preference alignment with progressive pertur- bation consistently outperforms SFT, achieving significant gains in translation quality with an average BLEU increase of 1.84 to 2.47 and CHRF increase of 2.85 to 4.01 compared to SFT for selected languages, while using the same positive training samples and under simi- lar computational constraints. This highlights the potential of our negative sample genera- tion strategy within KTO, especially in low- resource scenarios.

Abstract

Automatic detection and classification of stuttering events remain a challenging problem in speech processing, particularly due to the variability of disfluency types and the scarcity of large annotated datasets. This paper presents a signal processing-based approach for stuttering classification that avoids the data-intensive requirements of machine learning models. Our method focuses on syllable-level analysis using key acoustic parameters such as energy patterns, pitch contours, spectral stability, and temporal features, implemented through rule-based classification with predefined thresholds. The methodology involves automatic syllable segmentation followed by rule-based classification using thresholds on acoustic features. The dataset comprises annotated read and spontaneous speech from 106 Kannada-speaking adults, including 26 new participants evaluated during clinical trials. Three certified Speech-Language Pathologists (SLPs) conducted perceptual evaluations, and system outputs were validated using intraclass correlation (ICC), yielding strong agreement (textgreater 0.82) across stuttering types. The system achieved classification accuracies of 89% for blocks, 83% for repetitions, and 81% for prolongations. However, the system has several limitations: it relies on static threshold values, exhibits sensitivity to noise, and lacks automated severity grading. Additionally, generalizability across languages and speaking conditions requires further exploration. Machine learning models such as SVM, TDNN, and LSTM were trained and achieved a performance of around 30 to 60% accuracy. Our approach offers high interpretability, low computational cost, and real-time feasibility compared to machine learning methods. Future work will focus on adaptive thresholding, noise-robust processing, and hybrid approaches integrating signal processing with lightweight learning-based refinement.

Abstract

Natural walking offers high immersion in Virtual Reality (VR) but is constrained by physical space. Overlapping architectures, which reuse physical space, are a potential solution. Building on foundational principles of Architecturally Consistent Maze Generation in VR [1], we developed,implemented, and evaluated the Procedural Overlapping Maze System (POMS). POMS procedurally generates and prunes right-angled, overlapping maze corridors in real-time, enabling continuous natural walking in footprints ≥ 4x4m. This paper presents a controlled, between-group empirical evaluation (N=17 matched pairs) comparing POMS against a spatially equivalent static maze in an 8x8m tracked area using an Oculus Quest 3. Participants performed a time-constrained navigation task. We investigated effects on user engagement (time-on-task), usability (User Experience Questionnaire - UEQ), presence (igroup Presence Questionnaire - iPQ), and cybersickness (Simulator Sickness Questionnaire - SSQ). We hypothesized POMS would lead to increased time-on-task (H1), and that its dynamic nature would not negatively impact user experience, aiming for UEQ, iPQ, and SSQ outcomes comparable to a static environment (H2-H4). Results indicated no statistically significant differences in UEQ (H2) or iPQ (H3) between conditions, supporting our hypotheses that POMS could maintain these experiential qualities. Surprisingly, regarding cybersickness (H4), the increase in Oculomotor, Disorientation, and Total SSQ symptoms was significantly smaller in the POMS condition compared to the Static condition, while there was no significant difference for Nausea. This finding demonstrates an unexpected and substantial comfort benefit with POMS, exceeding our initial hypothesis of mere comparability. While POMS did not show a statistically significant increase in time-on-task (H1) in this study, a numerical trend suggesting longer engagement was observed. The markedly improved comfort achieved with POMS strongly suggests that such architectures have the potential to facilitate longer interaction periods, warranting further investigation into task design to fully leverage this benefit. These findings highlight that procedurally overlapping architectures like POMS can enable extended natural walking, not only without degrading core experiential qualities,but also by significantly enhancing user comfort.

Abstract

Toxicity identification in online multimodal environments remains a challenging task due to the complexity of contextual connections across modalities (e.g., textual and visual). In this paper, we propose a novel framework that integrates Knowledge Distillation (KD) from Large Visual Language Models (LVLMs) and knowledge infusion to enhance the performance of toxicity detection in hateful memes. Our approach extracts sub-knowledge graphs from ConceptNet, a large-scale commonsense Knowledge Graph (KG) to be infused within a compact VLM framework. The relational context between toxic phrases in captions and memes, as well as visual concepts in memes enhance the model's reasoning capabilities. Experimental results from our study on two hate speech benchmark datasets demonstrate superior performance over the state-of-the-art baselines across AU-ROC, F1, and Recall with improvements of 1.1%, 7%, and 35%, respectively. Given the contextual complexity of the toxicity detection task, our approach showcases the significance of learning from both explicit (i.e. KG) as well as implicit (i.e. LVLMs) contextual cues incorporated through a hybrid neurosymbolic approach. This is crucial for real-world applications where accurate and scalable recognition of toxic content is critical for creating safer online environments.

Abstract

Large Language Models (LLMs) have demonstrated remarkable capabilities in natural language understanding and generation. However, their tendency to memorize training data raises concerns regarding privacy, copyright compliance, and security, particularly in cases involving Personally Identifiable Information (PII). Effective machine unlearning techniques are essential to mitigate these risks, yet existing methods remain underdeveloped for LLMs due to their open-ended output space. In this work, we apply the Adaptive Representation Misdirection Unlearning (RMU) technique to unlearn sensitive information from LLMs. Through extensive experiments, we analyze the effects of unlearning across different decoder layers to determine the most effective regions for sensitive information removal. Our technique ranked 4th on the official leaderboard of both 1B parameter and 7B parameter models.

Abstract

Cancer metabolism is characterized by significant heterogeneity, presenting challenges for treatment efficacy and patient outcomes. Understanding this heterogeneity and its regulatory mechanisms at single-cell resolution is crucial for developing personalized therapeutic strategies. In this study, we employed a single-cell network approach to characterize malignant heterogeneity in gynecologic and breast cancers, focusing on the transcriptional regulatory mechanisms driving metabolic alterations. By leveraging single-cell RNA sequencing (scRNA-seq) data, we assessed the metabolic pathway activities and inferred cancer-specific protein-protein interactomes (PPI) and gene regulatory networks (GRNs). We explored the crosstalk between these networks to identify key alterations in metabolic regulation. Clustering cells by metabolic pathways revealed tumor heterogeneity across cancers, highlighting variations in oxidative phosphorylation, glycolysis, cholesterol, fatty acid, hormone, amino acid, and redox metabolism. Our analysis identified metabolic modules associated with these pathways, along with their key transcriptional regulators. These findings provide insights into the complex interplay between metabolic rewiring and transcriptional regulation in gynecologic and breast cancers, paving the way for potential targeted therapeutic strategies in precision oncology. Furthermore, this pipeline for dissecting coregulatory metabolic networks can be broadly applied to decipher metabolic regulation in any disease at single-cell resolution.

Abstract

Free-form electronics are one of the major chal- lenges being tackled in the present era. Along with the substrate and components, the conductive materials used in these also have to endure the applied strain, hence, thin film based conductors are being studied extensively. In this work, we report bending behavior of conductive CNT-PDMS thin film channels. Transmission Line Method (TLM) based resistance calculations have been done to precisely analyze the CNT thin film sheet resistance and the contact resistance. This exercise allows us to look at the effectiveness of CNT channels as a conductive path and also helps us understand their behavior over concentration, path length and strain caused due to bending. We report sheet resistance values of 254 Ω/□ and 1.41 Ω/□ for thin films of two different CNT concentrations. The contact resistances for these films were 230Ω and 315 Ω respectively. We observed that bending induced strain caused the resistance values to increase from 3.95 kΩ for flat to 13.1 kΩ for bending radius of 1.2 cm.

Abstract

Recent literature in soft robotics has been focused on bio-mimetic technologies, i.e., trying to replicate naturally-occurring structures, such as artificial muscles. Flapping structures have been reported, wherein a soft wing is paired with a solid actuator or vice-versa, but a fully soft actuator and wing setup has only recently been developed. In this work, we present the fabrication and characterization of the electrical response of a fully soft DEA-based flapping wing actuator. The actuator was fabricated using a urethane acrylate polymer elastomer with carbon nanotube (CNT) based electrodes. The 40 mm × 20 mm rectangular actuator was powered through two electrodes. It was subjected to static DC voltages in the range of 1 kV to 2 kV and the current and energy drawn by the actuator were observed. We report the average settling current varying from 7 μA to 43 μA and energy draw varying from 220 mJ to 715 mJ. Further tests on a similarly fabricated actuator revealed the possibility of Time Dependent Dielectric breakdown (TDDB) encountered at 2kV after nearly 120 minutes.