arXiv subjects

Tianyi Lyu

Publications and source records attributed to Tianyi Lyu.

14 recordsLinked to original sources

arXiv2026

ActiveFly-Bench: Aligning Embodied Question Answering with Vision-Language-Action for Aerial Embodied Perception

We introduce ActiveFly-Bench, the first benchmark to bridge cyberspace reasoning and physical-world interaction for UAV embodied perception. The benchmark decomposes active perception into three hierarchical tasks: Aerial Embodied Question Answering (Air-EQA), Observation Behavior Planning (OBP), and Fine-grained Language-guided UAV Control (FLUC), explicitly connecting high-level task understanding, behavior planning, and low-level control. The datasets are collected from both real-world and simulated outdoor environments for training and evaluation. We further develop ActiveFly, a closed-loop UAV agent that integrates visual-language reasoning with fine-grained control, and deploy it on a physical UAV platform. Experiments with representative VLMs and VLA models show that current UAV agents still struggle with behavior planning, viewpoint adjustment, and robust task completion in active perception. These results establish ActiveFly-Bench as a new testbed for embodied aerial intelligence.

cs.RO↗

arXiv2026

AB-Sync: Attention-Based Slot-Level Clock Synchronization Method for UWB-TDOA Localization Networks

Ultra-wideband (UWB) time-difference-of-arrival (TDOA) localization networks provide high-update-rate indoor location services for IoT and cyber-physical applications, but their accuracy depends on nanosecond-level clock synchronization among anchors. Existing wireless clock synchronization (WCS) methods typically estimate clock states at the synchronization-stage or interval level, whereas TDMA-based UWB-TDOA systems localize tags from blinks transmitted in discrete short slots inside each synchronization stage. We identify this granularity mismatch as a source of residual TDOA error and present AB-Sync, an attention-based slot-level clock synchronization method. AB-Sync models the relationship between the slot-specific clock-speed ratio required by a target tag blink and neighboring clock-fluctuation observations, thereby enabling tag-slot-level timestamp mapping without adding extra UWB synchronization messages. On a real UWB-TDOA testbed, AB-Sync reduces the multi-anchor average TDOA ranging STD.V by 9.4% and improves representative static localization accuracy by 18.6% compared with Deferred+3S-KF, the leading low-overhead baseline in our evaluation. In a five-slot multi-tag experiment, AB-Sync consistently improves localization stability across all TDMA slots, reducing STD.V by 5.3% on average and up to 16.2% per slot with no extra UWB synchronization overhead.

cs.NI↗

arXiv2026

What and When to Distill: Selective Hindsight Distillation for Multi-Turn Agents

Reinforcement learning can train LLM agents from sparse task rewards, but long-horizon credit assignment remains challenging: a single success-or-failure signal must be distributed across many actions. Existing methods rely on trajectory-level rewards or proxy signals, without fully leveraging per-step environmental feedback. Multi-turn agent settings are underexplored, where feedback can include error messages, page changes, observations, or reference trajectories. We systematically study five feedback sources and two insertion granularities and introduce SERL, a selective environment-reweighted learning framework. SERL uses the task reward to determine update direction, while environment feedback adjusts placement and magnitude, focusing on critical actions. On ALFWorld and WebShop, SERL achieves 90.0% and 80.1% success, outperforming strong RL and distillation baselines. Analysis shows that grounded, action-relevant feedback at meaningful points consistently outperforms indiscriminate use of longer or richer context.

cs.AI↗

arXiv2026

Beyond Mode Collapse: Distribution Matching for Diverse Reasoning

On-policy reinforcement learning methods like GRPO suffer from mode collapse: they exhibit reduced solution diversity, concentrating probability mass on a single solution once discovered and ceasing exploration of alternative strategies. We show this stems from reverse KL minimization's mode-seeking behavior, which reinforces the first high-reward trajectory found rather than maintaining a distribution over multiple diverse solutions. We propose DMPO (Distribution-Matching Policy Optimization), which prevents mode collapse through principled approximation of forward KL minimization. DMPO constructs a group level target distribution over sampled trajectories proportional to their rewards, then aligns the policy distribution to this target. This provides mode-covering behavior without requiring sampling from the intractable global target distribution, enabling sustained exploration throughout training. We validate DMPO on NP-hard combinatorial optimization, where exponentially many feasible solutions exist but only a few approach optimality, an ideal testbed for evaluating exploration. DMPO achieves 43.9% Quality Ratio on text-based NP-Bench (vs. GRPO's 40.1%) and 43.1% on vision-based NP-Bench (vs. 38.4%), demonstrating 9% and 12% relative improvements respectively. These gains generalize to mathematical reasoning (+2.0%) and out-of-domain tasks (+2.3%), showing that diversity-preserving training enhances general reasoning capabilities across modalities. Our work establishes distribution matching as a practical, principled approach to preventing mode collapse in on-policy RL, with consistent quality improvements demonstrating sustained exploration across diverse reasoning tasks.

cs.AI↗

arXiv2026

Escaping the Context Bottleneck: Active Context Curation for LLM Agents via Reinforcement Learning

Large Language Models (LLMs) struggle with long-horizon tasks due to the "context bottleneck" and the "lost-in-the-middle" phenomenon, where accumulated noise from verbose environments degrades reasoning over multi-turn interactions. To address this issue, we introduce a symbiotic framework that decouples context management from task execution. Our architecture pairs a lightweight, specialized policy model, ContextCurator, with a powerful frozen foundation model, TaskExecutor. Trained via reinforcement learning, ContextCurator actively reduces information entropy in the working memory. It aggressively prunes environmental noise while preserving reasoning anchors, that is, sparse data points that are critical for future deductions. On WebArena, our framework improves the success rate of Gemini-3.0-flash from 36.4% to 41.2% while reducing token consumption by 8.8% (from 47.4K to 43.3K). On DeepSearch, it achieves a 57.1% success rate, compared with 53.9%, while reducing token consumption by a factor of 8. Remarkably, a 7B ContextCurator matches the context management performance of GPT-4o, providing a scalable and computationally efficient paradigm for autonomous long-horizon agents.

cs.AI↗

arXiv2026

COINBench: Moving Beyond Individual Perspectives to Collective Intent Understanding

Understanding human intent is a high-level cognitive challenge for Large Language Models (LLMs), requiring sophisticated reasoning over noisy, conflicting, and non-linear discourse. While LLMs excel at following individual instructions, their ability to distill Collective Intent - the process of extracting consensus, resolving contradictions, and inferring latent trends from multi-source public discussions - remains largely unexplored. To bridge this gap, we introduce COIN-BENCH, a dynamic, real-world, live-updating benchmark specifically designed to evaluate LLMs on collective intent understanding within the consumer domain. Unlike traditional benchmarks that focus on transactional outcomes, COIN-BENCH operationalizes intent as a hierarchical cognitive structure, ranging from explicit scenarios to deep causal reasoning. We implement a robust evaluation pipeline that combines a rule-based method with an LLM-as-the-Judge approach. This framework incorporates COIN-TREE for hierarchical cognitive structuring and retrieval-augmented verification (COIN-RAG) to ensure expert-level precision in analyzing raw, collective human discussions. An extensive evaluation of 20 state-of-the-art LLMs across four dimensions - depth, breadth, informativeness, and correctness - reveals that while current models can handle surface-level aggregation, they still struggle with the analytical depth required for complex intent synthesis. COIN-BENCH establishes a new standard for advancing LLMs from passive instruction followers to expert-level analytical agents capable of deciphering the collective voice of the real world. See our project page on COIN-BENCH.

cs.IR↗

arXiv2025

Radiation-induced Instability of Organic-Inorganic Halide Perovskite Single Crystals

Organic-inorganic halide perovskites (OIHPs) are promising optoelectronic materials, but their instability under radiation environments restricts their durability and practical applications. Here we employ electron and synchrotron X-ray beams, individually, to investigate the radiation-induced instability of two types of OIHP single crystals (FAPbBr3 and MAPbBr3). Under the electron beam, we observe that 3-point star-style cracks grow on the surface of FAPbBr3, and bricklayer-style cracks are formed on the surface of MAPbBr3. Under the X-ray beam, a new composition without organic components appears in both FAPbBr3 and MAPbBr3. Such cracking and composition changes are attributed to the volatilization of organic components. We propose a volume-strain-based mechanism, in which the energy conversion results from the organic cation loss. Using nanoindentation, we reveal that beam radiations reduce the Youngs modulus and increase the hardness of both OIHPs. This study provides valuable insights into the structural and mechanical stabilities of OIHP single crystals in radiation environments.

cond-mat.mtrl-sci↗

arXiv2024

Electronic Health Records-Based Data-Driven Diabetes Knowledge Unveiling and Risk Prognosis

In the healthcare sector, the application of deep learning technologies has revolutionized data analysis and disease forecasting. This is particularly evident in the field of diabetes, where the deep analysis of Electronic Health Records (EHR) has unlocked new opportunities for early detection and effective intervention strategies. Our research presents an innovative model that synergizes the capabilities of Bidirectional Long Short-Term Memory Networks-Conditional Random Field (BiLSTM-CRF) with a fusion of XGBoost and Logistic Regression. This model is designed to enhance the accuracy of diabetes risk prediction by conducting an in-depth analysis of electronic medical records data. The first phase of our approach involves employing BiLSTM-CRF to delve into the temporal characteristics and latent patterns present in EHR data. This method effectively uncovers the progression trends of diabetes, which are often hidden in the complex data structures of medical records. The second phase leverages the combined strength of XGBoost and Logistic Regression to classify these extracted features and evaluate associated risks. This dual approach facilitates a more nuanced and precise prediction of diabetes, outperforming traditional models, particularly in handling multifaceted and nonlinear medical datasets. Our research demonstrates a notable advancement in diabetes prediction over traditional methods, showcasing the effectiveness of our combined BiLSTM-CRF, XGBoost, and Logistic Regression model. This study highlights the value of data-driven strategies in clinical decision-making, equipping healthcare professionals with precise tools for early detection and intervention. By enabling personalized treatment and timely care, our approach signifies progress in incorporating advanced analytics in healthcare, potentially improving outcomes for diabetes and other chronic conditions.

cs.LG↗

arXiv2024

Optimized CNNs for Rapid 3D Point Cloud Object Recognition

This study introduces a method for efficiently detecting objects within 3D point clouds using convolutional neural networks (CNNs). Our approach adopts a unique feature-centric voting mechanism to construct convolutional layers that capitalize on the typical sparsity observed in input data. We explore the trade-off between accuracy and speed across diverse network architectures and advocate for integrating an $\mathcal{L}_1$ penalty on filter activations to augment sparsity within intermediate layers. This research pioneers the proposal of sparse convolutional layers combined with $\mathcal{L}_1$ regularization to effectively handle large-scale 3D data processing. Our method's efficacy is demonstrated on the MVTec 3D-AD object detection benchmark. The Vote3Deep models, with just three layers, outperform the previous state-of-the-art in both laser-only approaches and combined laser-vision methods. Additionally, they maintain competitive processing speeds. This underscores our approach's capability to substantially enhance detection performance while ensuring computational efficiency suitable for real-time applications.

cs.CV↗

arXiv2024

IoT-Based 3D Pose Estimation and Motion Optimization for Athletes: Application of C3D and OpenPose

This study proposes the IoT-Enhanced Pose Optimization Network (IE-PONet) for high-precision 3D pose estimation and motion optimization of track and field athletes. IE-PONet integrates C3D for spatiotemporal feature extraction, OpenPose for real-time keypoint detection, and Bayesian optimization for hyperparameter tuning. Experimental results on NTURGB+D and FineGYM datasets demonstrate superior performance, with AP$^p50$ scores of 90.5 and 91.0, and mAP scores of 74.3 and 74.0, respectively. Ablation studies confirm the essential roles of each module in enhancing model accuracy. IE-PONet provides a robust tool for athletic performance analysis and optimization, offering precise technical insights for training and injury prevention. Future work will focus on further model optimization, multimodal data integration, and developing real-time feedback mechanisms to enhance practical applications.

cs.CV↗

arXiv2024

EITNet: An IoT-Enhanced Framework for Real-Time Basketball Action Recognition

Integrating IoT technology into basketball action recognition enhances sports analytics, providing crucial insights into player performance and game strategy. However, existing methods often fall short in terms of accuracy and efficiency, particularly in complex, real-time environments where player movements are frequently occluded or involve intricate interactions. To overcome these challenges, we propose the EITNet model, a deep learning framework that combines EfficientDet for object detection, I3D for spatiotemporal feature extraction, and TimeSformer for temporal analysis, all integrated with IoT technology for seamless real-time data collection and processing. Our contributions include developing a robust architecture that improves recognition accuracy to 92\%, surpassing the baseline EfficientDet model's 87\%, and reducing loss to below 5.0 compared to EfficientDet's 9.0 over 50 epochs. Furthermore, the integration of IoT technology enhances real-time data processing, providing adaptive insights into player performance and strategy. The paper details the design and implementation of EITNet, experimental validation, and a comprehensive evaluation against existing models. The results demonstrate EITNet's potential to significantly advance automated sports analysis and optimize data utilization for player performance and strategy improvement.

cs.CV↗

arXiv2024

Real-time Monitoring of Lower Limb Movement Resistance Based on Deep Learning

Real-time lower limb movement resistance monitoring is critical for various applications in clinical and sports settings, such as rehabilitation and athletic training. Current methods often face limitations in accuracy, computational efficiency, and generalizability, which hinder their practical implementation. To address these challenges, we propose a novel Mobile Multi-Task Learning Network (MMTL-Net) that integrates MobileNetV3 for efficient feature extraction and employs multi-task learning to simultaneously predict resistance levels and recognize activities. The advantages of MMTL-Net include enhanced accuracy, reduced latency, and improved computational efficiency, making it highly suitable for real-time applications. Experimental results demonstrate that MMTL-Net significantly outperforms existing models on the UCI Human Activity Recognition and Wireless Sensor Data Mining Activity Prediction datasets, achieving a lower Force Error Rate (FER) of 6.8% and a higher Resistance Prediction Accuracy (RPA) of 91.2%. Additionally, the model shows a Real-time Responsiveness (RTR) of 12 milliseconds and a Throughput (TP) of 33 frames per second. These findings underscore the model's robustness and effectiveness in diverse real-world scenarios. The proposed framework not only advances the state-of-the-art in resistance monitoring but also paves the way for more efficient and accurate systems in clinical and sports applications. In real-world settings, the practical implications of MMTL-Net include its potential to enhance patient outcomes in rehabilitation and improve athletic performance through precise, real-time monitoring and feedback.

eess.SP↗

arXiv2024

Accurate Inverse Process Optimization Framework in Laser Directed Energy Deposition

In additive manufacturing (AM), particularly for laser-based metal AM, process optimization is crucial to the quality of products and the efficiency of production. The identification of optimal process parameters out of a vast parameter space, however, is a daunting task. Despite advances in simulations, the process optimization for specific materials and geometries is developed through a time-consuming trial-and-error approach, which often lacks the versatility to address multiple optimization objectives. Machine learning (ML) provides a powerful tool to accelerate the optimization process, but most current studies focus on simple single-track prints, which hardly translate to manufacturing 3D components for engineering applications. In this study, we develop an Accurate Inverse process optimization framework in laser Directed Energy Deposition (AIDED), based on machine learning models and a genetic algorithm, to aid process optimization in laser DED processes. Using the AIDED, we demonstrate the following: (i) Accurately predict single-track (R2 score 0.995), multi-track (R2 score 0.969), and multi-layer (1.07% and 10.75% error in width and height, respectively) cross-sectional melt pool geometries directly from process parameters; (ii) Determine appropriate hatch spacing and layer thickness for fabricating fully dense (density > 99.9%) multi-track and multi-layer prints; (iii) Inversely identify optimal process parameters directly from customizable application objectives within 1-3 hours. We also validate the effectiveness of the AIDED experimentally by achieving two exemplary targets: fast print speed and fine print resolution. Furthermore, we show the high transferability of the framework from stainless steel to pure nickel. With AIDED, we pave a new way for ''aiding'' the process optimization in the laser-based AM processes that is applicable to a wide range of materials.

cond-mat.mtrl-sci↗

arXiv2023

Tailoring the mechanical properties of 3D microstructures: a deep learning and genetic algorithm inverse optimization framework

Materials-by-design has been historically challenging due to complex process-microstructure-property relations. Conventional analytical or simulation-based approaches suffer from low accuracy or long computational time and poor transferability, further limiting their applications in solving the inverse material design problem. Here, we establish a deep learning and genetic algorithm framework that integrates forward prediction and inverse exploration. This framework provides an end-to-end solution to achieve application-specific mechanical properties by microstructure optimization. In this study, we select the widely used Ti-6Al-4V to demonstrate the effectiveness of this framework by tailoring its microstructure and achieving various yield strength and elastic modulus across a large design space, while minimizing the stress concentration factor. Compared with conventional methods, our framework is efficient, versatile, and readily transferrable to other materials and properties. Paired with additive manufacturing's potential in controlling local microstructural features, our method has far-reaching potential for accelerating the development of application-specific, high-performing materials.

cond-mat.mtrl-sci↗