Contactless respiratory monitoring from video is a promising alternative to wearable sensors but remains sensitive to motion artifacts and variations in viewpoint and lighting, which degrade the reliability of respiratory rate (RR) estimates. Traditional signal processing methods often struggle to separate physiological motion from visual artifacts, while large multimodal models (LMMs) can reason about these artifacts but are often too computationally demanding for privacy-sensitive or embedded deployments. This work aims to improve robustness and deployability by explicitly decoupling visual perception from downstream reasoning. We propose VLS-RR (Visual–Language– SLM Respiratory Rate) auditor, a three-stage framework where video segments are converted into fused motion signals and summary plots. A vision–language model (VLM) then produces a textual description of rhythm, breath count, and artifacts. Finally, small language models (SLMs) perform chain-of-thought (CoT) auditing over this evidence and the signal-derived RR candidates. We evaluate VLS-RR on a 50-video belt-synchronized dataset spanning diverse breathing patterns, viewpoints, and lighting conditions. Compared with signal-only baselines, VLS-RR reduces segment-level MAE by ≈ 30% and improves R 2 from 0.85 to 0.92. It also outperforms numeric-only SLM baselines with an additional ≈ 25% MAE reduction. Finally, on an embedded edge device, compact SLM auditors run with 0.82–1.71 s latency per 6 s window, indicating that decoupling perception and reasoning enables accurate, resource-efficient RR estimation from video.

In mission-critical scenarios, autonomous agents often operate in GPS-denied or GPS-degraded environments, making it chal- lenging to localize agents relative to themselves and objects of interest (e.g., potential cover) or adversarial robots. While vision- and radio- based modalities have individually been used to estimate relative ranges between agents and surrounding objects, each modality suffers from in- herent limitations. Vision-based range estimation degrades significantly under low image overlap, partial visibility, or when an agent is too close to an object, resulting in effective zoom-in and loss of geometric context. Conversely, radio-based ranging is susceptible to multipath interference, signal fading, and environmental variability, which can substantially reduce range accuracy and reliability. To address these challenges, we introduce CAViAR, a quality-aware multimodal fusion framework for ac- curate relative range estimation among collaborative autonomous agents. CAViAR assigns modality-specific reliability scores and performs sta- tistical fusion to adaptively weigh vision and radio inputs based on their estimated quality. The framework employs modality-specific quality estimators augmented with temporal features and integrates MBConv blocks to enable efficient feature processing on resource-constrained robotic platforms. We validate CAViAR on ROSbot 2 and ROSbot 2 Pro platforms using an in-house dataset collected across diverse indoor and outdoor environments. Experimental results demonstrate that our approach outperforms single-modality baselines by approximately 21% over vision-only and 36% over radio-only range estimates. Moreover, CAViAR adapts robustly to variations in scene structure, viewpoint overlap, and occlusions without requiring fine-tuning on new environments, highlighting its practicality for real-world deployment.

Large deep neural networks (DNNs), especially transformer-based and multimodal architectures, are computationally demanding and challenging to deploy on resourceconstrained edge platforms like field robots. These challenges intensify in mission-critical scenarios (e.g., disaster response), where robots must collaborate under tight constraints on bandwidth, latency, and battery life, often without infrastructure or server support. To address these limitations, we present COHORT, a collaborative DNN inference and taskexecution framework for multi-robot systems built on the Robotic Operating System (ROS). COHORT employs a hybrid offline–online reinforcement learning (RL) strategy to dynamically schedule and distribute DNN module execution across robots. Our key contributions are threefold: (a) Offline RL policy learning combined with Advantage-Weighted Regression (AWR), trained on auction-based task allocation data from heterogeneous DNN workloads across distributed robots, (b) Online policy adaptation via Multi-Agent PPO (MAPPO), initialized from the offline policy and fine-tuned in real time, and (c) comprehensive evaluation of COHORT on vision-language model (VLM) inference tasks such as CLIP and SAM, analyzing scalability with increasing robot/workload and robustness under . We benchmark COHORT against genetic algorithms and multiple RL baselines. Experimental results demonstrate that COHORT reduces battery consumption by 15.4%, and increases GPU utilization by 51.67%, while satisfying framerate and deadline constraints 2.55x of the time.

Continuous, non-invasive respiratory rate (RR) monitoring is essential for the early diagnosis of many medical problems. However, conventional contact-based sensors frequently under-perform in dynamic situations that can be uncomfortable and require human intervention. To overcome these limitations, we propose E2RespUNet, an end-to-end system that uses multimodal video data to estimate breathing rates and reconstruct respiratory signals using an attention-enhanced U-Net architecture. Our method combines optical flow analysis with preprocessing, detrending, and normalization to reliably extract chest motion features in a variety of settings. According to our study in both the temporal and frequency domains, E2RespUNet surpasses existing baseline models by lowering the mean absolute error by up to 21 % in the sleep dataset and by up to 28% in the in-house dataset.

Integrating multimodal data such as RGB and LiDAR from multiple views significantly increases computational and communication demands, which can be challenging for resource-constrained autonomous agents while meeting the time-critical deadlines required for various mission-critical applications. To address this challenge, we propose CoOpTex, a collaborative task execution framework designed for cooperative perception in distributed autonomous systems (DAS).

The usage of Lithium-ion (Li-ion) batteries has gained widespread popularity across various industries, from powering portable electronic devices to propelling electric vehicles and supporting energy storage systems. A central challenge in Li-ion battery reliability lies in accurately predicting their Remaining Useful Life (RUL), which is a critical measure for proactive maintenance and predictive analytics. This study presents a novel approach that harnesses the power of multiple denoising modules, each trained to address specific types of noise commonly encountered in battery data. Specifically, a denoising auto-encoder and a wavelet denoiser are used to generate encoded/decomposed representations, which are subsequently processed through dedicated self-attention transformer encoders. After extensive experimentation on NASA and CALCE data, a broad spectrum of health indicator values are estimated under a set of diverse noise patterns. The reported error metrics on these data are on par with or better than the state-of-the-art reported in recent literature.