Abstract

Scene understanding and reasoning has been a fundamental problem in 3D computer vision, requiring models to identify objects, their properties, and spatial or comparative relationships among the objects. Existing approaches enable this by creating scene graphs using multiple inputs such as 2D images, depth maps, object labels, and annotated relationships from specific reference view. However, these methods often struggle with generalization and produce inaccurate spatial relationships like "left/right", which become inconsistent across different viewpoints. To address these limitations, we propose Viewpoint-Invariant ZerO-shot scene graph generation for 3D scene Reasoning (VIZOR). VIZOR is a training-free, end-to-end framework that constructs dense, viewpoint-invariant 3D scene graphs directly from raw 3D scenes. The generated scene graph is unambiguous, as spatial relationships are defined relative to each object’s front-facing direction, making them consistent regardless of the reference view. Furthermore, it infers open-vocabulary relationships that describe spatial and proximity relationships among scene objects without requiring annotated training data. We conduct extensive quantitative and qualitative evaluations to assess the effectiveness of VIZOR on scene graph generation and downstream tasks, such as query-based object grounding. VIZOR outperforms state-of-the-art methods, showing clear improvements in scene graph generation and achieving 22% and 4.81% gains in zero-shot grounding accuracy on the Replica and Nr3D datasets, respectively.

Abstract

Generating realistic human motions that naturally respond to both spoken language and physical objects is crucial for interactive digital experiences. Current methods, however, address speech-driven gestures or object interactions independently, limiting real-world applicability due to a lack of integrated, comprehensive datasets. To overcome this, we introduce InteracTalker, a novel framework that seamlessly integrates prompt-based object-aware interactions with co-speech gesture generation. We achieve this by employing a multi-stage training process to learn a unified motion, speech, and prompt embedding space. To support this, we curate a rich human-object interaction dataset, formed by augmenting an existing text-to-motion dataset with detailed object interaction annotations. Our framework utilizes a Generalized Motion Adaptation Module that enables independent training, adapting to the corresponding motion condition, which is then dynamically combined during inference. To address the imbalance between heterogeneous conditioning signals, we propose an adaptive fusion strategy, which dynamically reweights the conditioning signals during diffusion sampling. InteracTalker successfully unifies these previously separate tasks, outperforming prior methods in both co-speech gesture generation and object-interaction synthesis, outperforming gesture-focused diffusion methods, yielding highly realistic, object-aware full-body motions with enhanced realism, flexibility, and control.(https://sreeharirajan.github.io/projects/InteracTalker/)

Abstract

Early-stage fruit yield prediction plays a key role in supporting timely agronomic decisions, enhancing market planning, and empowering farmers with data-driven insights. Over the years, most approaches to yield estimation have focused on fruit counting techniques, typically performed just before harvest. While these methods have proven useful, they often come into play late in the cultivation cycle, limiting their impact on early planning and resource optimization. In this work, we introduce a comprehensive baseline framework for predicting mango yield at an earlier stage - during flowering - using image-based learning. Our contributions are twofold. (i) Our approach combines a SegFormer-based segmentation model with a regression pipeline to estimate yield from images, while also exploring the role of contextual features such as weather and scale. (ii) This work introduces a novel benchmark and an enriched dataset, paving the way for scalable, automated tools that can assist farmers and stakeholders in making proactive decisions throughout the mango growing season. Our work demonstrates that for multi-modal yield prediction, integrating features that complement visual representations (like scale) can be more impactful than using features with a stronger standalone linear correlation (like weather). Our single-image model, based on the SegFormer-B1 encoder, achieved a mean absolute error (MAE) of 7.68, R² of 0.76, and mean squared error (MSE) of 115.48. These results highlight the promise of vision-based models for yield estimation from early-stage flowering cues. To the best of our knowledge, this is the first work to address the prediction of mango yield using images from the flowering stage and weather data.

Abstract

Temporal Graph Neural Networks (TGNNs) are increasingly used in high-stakes domains, such as financial forecasting, recommendation systems, and fraud detection. However, their susceptibility to poisoning attacks poses a critical security risk. We introduce LORETTA (Low Resource Twophase Temporal Attack), a novel adversarial framework on Continuous-Time Dynamic Graphs, which degrades TGNN performance by an average of 29.47% across 4 widely benchmark datasets and 4 State-of-the-Art (SotA) models. LORETTA operates through a two-stage approach: (1) sparsify the graph by removing high-impact edges using any of the 16 tested temporal importance metrics, (2) strategically replace removed edges with adversarial negatives via LORETTA’s novel degree-preserving negative sampling algorithm. Our plug-and-play design eliminates the need for expensive surrogate models while adhering to realistic unnoticeability constraints. LORETTA degrades performance by upto 42.0% on MOOC, 31.5% on Wikipedia, 28.8% on UCI, and 15.6%on Enron. LORETTA outperforms 11 attack baselines, remains undetectable to 4 leading anomaly detection systems, and is robust to 4 SotA adversarial defense training methods, establishing its effectiveness, unnoticeability, and robustness.

Abstract

Time-Sensitive Networking (TSN) is crucial for ensuring deterministic communication in real-time applications. In TSN, Network Traversal Time (NTT) is a critical metric that quantifies the timeliness of received data. Maintaining low NTT is essential for ensuring accurate system responses and to prevent outdated information from affecting real-time operations. However, the presence of non-zero arrival jitter negatively impacts NTT and poses challenges in providing predictable performance. When jitter increases, packets may arrive inconsistently, causing some to be delayed while others may arrive in bursts. To address this, we analyze the NTT bounds for time-critical flows, influenced by arrival jitter. We propose an analytical framework to estimate best and worst-case ST slots for data transmission, considering interference from higher priority flows and establishing data reachability along the network path across the switches. To validate our analysis, we conducted experiments using both a synthetic task set and an automotive use case. We also analyzed the impact of flow parameters on NTT determination, which, in some scenarios, led to pessimistic bound estimations.

Abstract

This paper presents a real-time, low-latency communication and energy- efficient framework for IoT-enabled micromobility systems, specifically targeting gated environments such as residential campuses, educational institutions and industrial zones. As a core application platform, an IoT-enabled Segway is deployed and integrated with the proposed framework, serving as a practical micro-mobility solution within gated communities. The proposed system enables both Vehicle-to-Vehicle (V2V) communication using ESP-NOW, and Vehicle-to-Infrastructureto-Vehicle (V2I2V) communication using Wi-Fi-based UDP, implemented using ESP32-S3 and NodeMCU microcontrollers integrated with IMU, ToF, Ultrasonic, and GPS sensors. To evaluate performance, end-to-end latency is measured across varying distances under both Line-of-Sight (LoS) and Non-Lineof-Sight (NLoS) conditions. Along with the latency measurement, power profiling system is also implemented using the INA226 sensor, comparing the energy consumption of ESP-NOW and 4G-based communication. To extend this analysis, Simulation of Urban MObility (SUMO) simulation was performed. Results demonstrate the effectiveness of the proposed framework in supporting reliable micro-mobility communication, offering a practical foundation for intelligent transport in localized smart environments. This work introduces a novel, energy-efficient, and infrastructure-independent V2X communication solution tailored for smart gated communities.

Abstract

The rapid evolution of vehicular edge computing (VEC) has made task offloading to roadside units (RSUs) essential for addressing the computational demands of modern vehicular applications. However, efficiently partitioning and scheduling tasks across RSUs in dynamic environments remains challenging. This paper introduces a novel algorithm, HAPTA (Horizontal, Adaptive Learning-based, Partitioned, Timeslot-based Algorithm), a time-slot based task partitioning and offloading approach using the Upper Confidence Bound (UCB) algorithm, to adaptively allocate tasks to RSUs. We consider task offloading horizontally across RSU nodes. Unlike static or heuristic-based scheduling methods, the proposed HAPTA framework dynamically selects RSUs with optimal resource availability while considering vehicular mobility and task deadlines. The framework supports splitting tasks into subtasks, enabling efficient resource utilization and meeting stringent latency constraints. Experimental evaluations on real-world vehicular datasets demonstrate that the HAPTA approach outperforms previously introduced heuristic-based methods, achieving higher task completion rates and lower latency, particularly in high-density scenarios.

Abstract

Automated Guided Vehicles (AGVs) are widely used for material handling across various industries, such as warehouses, manufacturing plants, and automated container terminals. As AGVs are battery-powered, it is necessary to schedule AGV's tasks considering energy-related aspects. That is, understanding how an AGV's energy is consumed or recharged in a given operational environment is crucial to optimizing both operational efficiency and energy consumption. This article presents a comprehensive analysis of energy-aware AGV operations, emphasizing three critical areas: the impact of the operational environment, energy consumption modeling, and energy-aware decision-making for scheduling and path planning. This article examines the layout structures and energy supply strategies in two different environments: automated container terminals and workshops, to understand the environmental impact on AGV energy efficiency. A comprehensive review of energy consumption models provides insight into the physical characteristics of AGVs' energy consumption. Finally, we provide variations of task scheduling and path planning problems that explicitly take into account such energy aspects and introduce how literature tackles to solve those problems.

Abstract

Efficient traffic signal control is critical for reducing urban congestion, yet traditional rule-based and machine learning approaches fail to adapt to dynamic conditions. Reinforcement Learning (RL) provides a data-driven alternative, and this study evaluates classical exploration strategies Epsilon-Greedy, Thompson Sampling, Upper Confidence Bound (UCB), and Softmax alongside hybrid variants including Softmax with Temperature Annealing (SMX-TA), Softmax with Entropy Regularization (SMX-ER), and their combinations with UCB and Thompson Sampling. Our framework is based on value-based RL (Double Deep Q-Networks), chosen for scalability and stability compared to on-policy methods such as PPO and A2C, and implemented with parallel simulation in SUMO and CityFlow across synthetic grids ( 1 × 1 , 4 × 4 , 6 × 6 ) and real-world datasets (Hyderabad, New York). Benchmarking against classical baselines (FixedTime, MaxPressure) and recent RL models (FRAP, CoLight, GCN) shows that the proposed SMX-ER+UCB consistently yields the lowest average travel times and robust adaptability across traffic conditions. An ablation study confirms the complementary benefits of entropy regularization and UCB, while integration into scalable controllers such as CoLight demonstrates network-level generalization. These results highlight hybrid exploration as an effective and practical strategy for real-time, city-wide traffic optimization. Such large-scale optimization requires high-performance computing (HPC) to support parallel simulations, GPU-accelerated training, and multi-agent coordination, underscoring the necessity of HPC frameworks for intelligent transportation systems.

Abstract

Psychomotor changes, while crucial indicators of depres- sion, remain underrepresented in clinical observations. We examined the relationship between depression and psychomotor behavior by analyzing head-tracking data related to yaw movements made during exploration of 360° emotional videos, alongside valence and arousal ratings on 9-point Likert scale. Symptoms of depres- sion were recorded using the Patient Health Question- naire (PHQ-9). While subjective ratings for valence and arousal showed no differences across depression groups, the head-tracking data revealed novel results. Individu- als with moderate to severe depression exhibited signif- icantly lower scanning speed and standard deviation in yaw movement compared to minimal to mild depression. Although preliminary, these results underscore the im- portance of psychomotor measures in diagnosis, risk as- sessment, and monitoring in psychiatric care, alongside subjective evaluations.