We present a visual encoding method for semantic mapping in indoor environments, designed to minimize redundancy in image data and support federated learning across a fleet of service robots. Our pipeline combines 2D LiDAR-based segmentation with RGB image filtering based on geometric orientation, distance, visibility, and uniqueness. The result is a compact set of representative visual samples suitable for downstream semantic tasks such as object recognition or language grounding. We evaluate our method in a Gazebo simulation using a TurtleBot platform and compare it against a naive odometry-based sampling strategy. Our approach achieves up to 57.5\% reduction in collected images while preserving scene coverage. Additionally, we demonstrate how multiple robots can collaboratively improve the visual map in a federated setup, reducing collection time and enabling model generalization across diverse environments. The proposed method offers an efficient and scalable solution for semantic mapping under bandwidth and computation constraints.
02683nas a2200133 4500000000100000000000100001008004100002100001800043700002000061700002100081245010200102856005500204520229000259 2026 d 1 aNur Kelesoglu 1 aŁukasz Sobczak 1 aJoanna Domańska 00 aWhen LLMs Speak ZigBee: Exploring Low-Latency and Reasoning Models for Network Traffic Generation uhttps://www.nature.com/articles/s41598-026-37246-y 3 aThe generation of realistic network traffic is a critical requirement for testing, simulation, and security evaluation in ZigBee-based IoT systems. In this study, we propose a novel framework that extracts sample ZigBee traffic packets and generates semantically meaningful and protocol-compliant synthetic traffic by using Large Language Models (LLMs) such as GPT-4.1 and GPT-5. Unlike traditional rule-based or statistical generators, our approach is data-driven and incorporates sample-based few-shot learning, prompt engineering, and a human-in-the-loop feedback mechanism. To evaluate the effectiveness of the proposed framework, we conduct two sets of experiments. The first focuses on generating unidirectional traffic to emulate typical device-to-hub communication, while the second extends this setup to bidirectional exchanges, capturing realistic request--response dynamics and interaction patterns. The realism of the generated traffic is assessed using a multi-dimensional evaluation framework that includes statistical similarity measures, such as Jensen--Shannon Divergence, as well as protocol compliance, semantic correctness, temporal consistency, and diversity metrics. In addition, we compare the performance of LLM-based generators with classical deep learning baselines, including recurrent neural networks (RNNs) and generative adversarial networks (GANs). We further analyze the computational cost and the impact of internal reasoning effort on traffic generation by systematically evaluating different GPT-5 reasoning configurations. Experimental results show that both GPT-4.1 and GPT-5 successfully learn the temporal and structural dependencies of ZigBee traffic and significantly outperform RNN and GAN baselines in terms of semantic correctness and long-duration generation. Across all experiments, GPT-4.1 consistently generates traffic that more closely resembles real ZigBee traffic while requiring substantially lower computational resources. These findings highlight that low-latency, non-reasoning LLMs can be particularly well suited for highly structured, protocol-constrained network traffic generation tasks, and demonstrate the potential of LLM-based approaches for realistic IoT traffic generation in research and security evaluation.
00560nas a2200109 4500000000100000008004100001260026300042100001800305700002000323700002100343245008600364 2026 d b(This work was partly supported by the National Centre for Research and Development (NCBR) and co-funded by the European Union under the European Funds for ModernEconomy(FENG)Programme(SMARTKonsorcja),projectno.FENG.01.01-IP.01-A0GV/24-00,) aKraków, Poland 1 aNur Kelesoglu 1 aŁukasz Sobczak 1 aJoanna Domańska 00 aCPU Utilization Forecasting in Large-Scale Cloud Clusters Using Gradient Boosting 01600nas a2200133 4500000000100000008004100001260004000042100002000082700001800102700002100120245007800141856005000219520119700269 2026 d c10.08.2026 bIEEE aKitakyushu, Japan 1 aŁukasz Sobczak 1 aNur Kelesoglu 1 aJoanna Domańska 00 aReliability-Aware LLM Reasoning: Handling Uncertainty in Robot Perception uhttps://ieeexplore.ieee.org/document/11714851 3 aRobots operating in human environments must often make decisions based on perceptual information that
is uncertain, incomplete, or ambiguous. This paper proposes a reliability-aware reasoning framework that enables large
language models (LLMs) to account for perceptual uncertainty when selecting actions in human-robot interaction scenarios.
The environment is represented as a structured scene composed of detected objects enriched with confidence estimates, attribute reliability, and spatial uncertainty information. Using this representation, the LLM evaluates candidate objects through a reliability scoring mechanism that integrates multiple sources of perceptual evidence and supports uncertainty-aware decision making.The proposed approach is evaluated using perception episodes with controlled levels of uncertainty and compared with a baseline LLM-based matching strategy that ignores perceptual
reliability. Experimental results show that incorporating uncertainty-aware reasoning substantially improves decision
robustness under medium and high uncertainty conditions while reducing safety-critical behaviors caused by overconfident
autonomous decisions.
Safe and reliable human–robot interaction remains a challenge, especially for Large Language Model (LLM)-powered socially assistive robots in elderly care. This paper proposes a risk-aware response framework that explicitly models the safety implications of user queries. We introduce a Query Risk Assessment Module (QRAM), which computes a Query Risk Score (QRS) using structured semantic indicators to classify user inputs into different risk levels. Based on this, we develop two strategies: risk-aware response refinement and risk-aware response generation. To assess safety, we introduce the Risk-Aware Response Safety Score (RRSS), a metric that captures both the presence of risk-bearing language and the absence of necessary safety-critical guidance. Experimental results across multiple LLMs demonstrate that incorporating query-level risk awareness reduces response-level risk, with both strategies outperforming raw responses.
00303nas a2200109 4500000000100000008004100001260001800042100002000060700001800080700002000098245007500118 2026 d aNaples, Italy 1 aŁukasz Sobczak 1 aNur Kelesoglu 1 aSławomir Nowak 00 aRisk-Aware Semantic Grounding for Trustworthy LLM-Based Robot Planning 00420nas a2200121 4500000000100000008004100001260003000042100001800072700002000090700002100110245011600131856005100247 2026 d c10/05/2026 aNaples, Italy 1 aNur Kelesoglu 1 aŁukasz Sobczak 1 aJoanna Domańska 00 aFrom Plots to Words: Model-Aware Multimodal Explanations as a Foundation for Accessible, Non-Visual Interaction uhttps://dl.acm.org/doi/10.1145/3776591.3833865 02216nas a2200169 4500000000100000008004100001100001700042700002500059700002000084700002400104700002300128700001800151700001600169700001900185245007600204520176600280 2026 d 1 aAygün Varol 1 aKatarzyna Kołodziej 1 aŁukasz Sobczak 1 aMichał Romaszewski 1 aPrzemysław Głomb 1 aNaser Motlagh 1 aMirka Leino 1 aJohanna Virkki 00 aEnabling Cloud-Level Accuracy in Edge AI through IoT Data Preprocessing 3 aLarge language models (LLMs) offer a natural-language interface for interpreting Internet of Things (IoT) sensor data in smart environments; however, cloud deployment introduces latency, privacy, and connectivity concerns. Local LLMs can reduce these limitations, but compact edge-deployable models often show weaker numerical reasoning when raw sensor readings are provided directly. This paper investigates whether prompt-side preprocessing can improve the accuracy--latency trade-off of local LLMs for environmental monitoring. We propose a structured prompt construction framework that transforms raw air-quality and thermal-comfort measurements into progressively enriched textual representations: raw sensor values, threshold-aware descriptions, and compact environmental summary flags. The approach is evaluated using indoor Raspberry Pi/BME680 datasets from Tampere University and outdoor air-quality datasets from Helsinki, Katowice, and Warsaw. We construct a binary LLM query dataset covering air quality, thermal comfort, and joint environmental conditions, and evaluate five local and five cloud LLMs across three prompt variants and two inference modes, with and without chain-of-thought prompting. The results show that prompt enrichment substantially improves local-model accuracy. In No-CoT mode, local accuracy increases from 50.9\% to 81.7\% indoors and from 63.7\% to 89.3\% outdoors from the raw to the most enriched prompt. Local No-CoT inference is the fastest configuration, with mean latency close to 0.22 s, while CoT substantially increases inference time. These findings suggest that lightweight prompt-side preprocessing can narrow the local--cloud performance gap and support low-latency IoT analytics in smart environments.
02266nas a2200181 4500000000100000008004100001100001700042700002500059700002000084700002400104700002300128700001800151700001600169700001900185245007600204856003800280520176600318 2026 d 1 aAygün Varol 1 aKatarzyna Kołodziej 1 aŁukasz Sobczak 1 aMichał Romaszewski 1 aPrzemysław Głomb 1 aNaser Motlagh 1 aMirka Leino 1 aJohanna Virkki 00 aEnabling Cloud-Level Accuracy in Edge AI through IoT Data Preprocessing uhttps://arxiv.org/abs/2606.22496 3 aLarge language models (LLMs) offer a natural-language interface for interpreting Internet of Things (IoT) sensor data in smart environments; however, cloud deployment introduces latency, privacy, and connectivity concerns. Local LLMs can reduce these limitations, but compact edge-deployable models often show weaker numerical reasoning when raw sensor readings are provided directly. This paper investigates whether prompt-side preprocessing can improve the accuracy--latency trade-off of local LLMs for environmental monitoring. We propose a structured prompt construction framework that transforms raw air-quality and thermal-comfort measurements into progressively enriched textual representations: raw sensor values, threshold-aware descriptions, and compact environmental summary flags. The approach is evaluated using indoor Raspberry Pi/BME680 datasets from Tampere University and outdoor air-quality datasets from Helsinki, Katowice, and Warsaw. We construct a binary LLM query dataset covering air quality, thermal comfort, and joint environmental conditions, and evaluate five local and five cloud LLMs across three prompt variants and two inference modes, with and without chain-of-thought prompting. The results show that prompt enrichment substantially improves local-model accuracy. In No-CoT mode, local accuracy increases from 50.9\% to 81.7\% indoors and from 63.7\% to 89.3\% outdoors from the raw to the most enriched prompt. Local No-CoT inference is the fastest configuration, with mean latency close to 0.22 s, while CoT substantially increases inference time. These findings suggest that lightweight prompt-side preprocessing can narrow the local--cloud performance gap and support low-latency IoT analytics in smart environments.
03010nas a2200181 4500000000100000008004100001100001700042700002500059700002000084700002400104700002300128700001800151700001600169700001900185245011600204856005200320520245600372 2026 d 1 aAygün Varol 1 aKatarzyna Kołodziej 1 aŁukasz Sobczak 1 aMichał Romaszewski 1 aPrzemysław Głomb 1 aNaser Motlagh 1 aMirka Leino 1 aJohanna Virkki 00 aAIRWISE: Environmental Sensor and LLM Query Dataset for Edge AI-Based Air-Quality and Thermal-Comfort Analytics uhttps://zenodo.org/doi/10.5281/zenodo.20783874 3 aAIRWISE is an open benchmark dataset for edge AI, IoT environmental monitoring, and large language model (LLM) evaluation on real sensor data. It combines indoor air-quality and thermal-comfort measurements, outdoor air-quality and meteorological time series, and a labeled binary question-answering benchmark for threshold-aware reasoning and anomaly detection. The dataset supports the study “Enabling Cloud-Level Accuracy in Edge AI through IoT Data Preprocessing”: https://doi.org/10.48550/arXiv.2606.22496
AIRWISE provides three core resources:
Indoor environmental sensor dataset (BME680, 1-minute resolution). 146,000+ minute-averaged readings from Raspberry Pi + Bosch BME680 nodes in three real micro-environments (office, kitchen, hallway) at Tampere, Finland (2025-11-14 to 2025-12-18): temperature, relative humidity, barometric pressure, gas resistance, indoor air quality (IAQ) proxy, z-score anomaly detectors, and limit-annotated long-format tables with ground-truth anomaly flags.
Outdoor air-quality and meteorological time series (hourly, full-year 2023). Helsinki (Finland), Katowice and Warsaw (Poland): NO₂, CO, O₃, PM₂.₅, temperature, and relative humidity assembled from national monitoring networks (FMI, GIOŚ, IMGW-PIB), as cleaned single-series city tables (≈8,760 hourly records per city) and limit-annotated long-format tables with anomaly flags.
Binary LLM query benchmark (1,440 labeled yes/no questions). 240 natural-language questions per site across six sites, balanced by task type (80 air quality, 80 thermal comfort, 80 joint environmental condition), each with ground-truth answer, violation label, offending factors, and rationale — ready for benchmarking local/edge and cloud LLMs on sensor-grounded reasoning.
Also included: reference air-quality and thermal-comfort limits (EU air-quality standards and Finnish/Polish national guidelines); the unmodified institutional source files for the outdoor component (FMI exports, GIOŚ archive workbooks and station metadata, IMGW-PIB synoptic tables); the complete dataset-generation scripts covering every pipeline step (raw sources → cleaned tables → limit-annotated data → LLM queries) with a requirements.txt for a verified minimal Python environment; SHA-256 checksums; and a dependency-free validation script. Schemas, exact row counts, coverage, and usage examples are documented in README.md and MANIFEST.md.