Om Rajendra
Kathalkar

Image-based air quality estimation offers a scalable alternative to sparse sensor-based monitoring. To address limitations of narrow field-of-view imagery, we introduce PAN-AQI — a large-scale dataset comprising 33,982 panoramic 360° street-view images collected over 18 days and 1000 km across Hyderabad and Secunderabad, India, with co-located PM₂.₅, PM₁₀, Temperature, and Humidity measurements. We further propose PANQIFormer, a multimodal transformer framework combining spatial-zone reasoning, attention-based multimodal fusion, and ordinal-aware learning for both AQI category classification and continuous AQI regression. PANQIFormer achieves consistent improvements over prior methods, reducing AQI MAE from 21.59 to 10.45. Together, PAN-AQI and PANQIFormer establish panoramic sensing as a promising direction for vision-based air-quality monitoring.

Industrial autonomous mobile robots increasingly depend on reliable wireless connectivity for real-time control and data streaming, yet existing path planning methods fail to account for the complex interplay between spatial geometries and material-dependent signal propagation. This paper presents a spatial-aware transformer framework for network quality prediction that explicitly encodes both geometric layouts and electromagnetic properties of building materials. Validation against the iV2I+ real-world dataset achieved R²=0.87 for SNR predictions. Evaluation across 8,840 test paths demonstrated SNR prediction RMSE of 2.31 dB (R²=0.891), a 19.5% improvement over state-of-the-art baselines. Zero-shot evaluation on mixed-material scenarios yielded only 10% performance degradation, eliminating facility-specific retraining.

AMRs in industrial environments require reliable wireless connectivity for coordination, control, and safety operations. This paper presents a framework for network-aware path planning incorporating wireless network quality metrics as path constraints. Sionna-based ray-tracing simulations are validated against real-world measurements (R²=0.87 for SNR, 0.82 for throughput). Three planners were implemented: A* with network constraints, CVAE-based neural path planning, and GNN-based multi-path planning. CVAE achieved 95.2% constraint satisfaction; GraphMP showed 23% shorter planning times.

AQIFormer is a transformer-based ensemble architecture addressing cross-city generalization in image-based air quality estimation. Through innovative dual-view integration, weather-aware attention mechanisms, and multi-task learning, it combines front and rear traffic imagery with meteorological parameters. Evaluated on 26,678 synchronized image pairs, achieving 89.96% accuracy — a 14.96% improvement over state-of-the-art. Cross-city generalization on an independent Nagpur dataset achieves 81.67% accuracy with only 8.29% performance degradation using few-shot adaptation.

TRAQID is a novel dataset of 26,678 synchronized front and rear traffic images with co-located weather parameters, PM₂.₅, PM₁₀ levels, and six-category AQI values. Collected over 70+ hours across Hyderabad and Secunderabad, India, spanning multiple seasons with diverse day/night imagery under unstructured traffic conditions. Establishes a challenging benchmark for image-based AQI estimation.

A protocol for using mobile search agents to identify PM₂.₅ emission hotspots in urban environments using IoT devices mounted on a mobile platform. Applied to Hyderabad, India, identifying short-range variability of PM₂.₅ using IoT sensing calibrated against a reference instrument. Random forest regression was most effective for calibration. The approach can be applied to any mobile platform — walkers, cyclists, drones, or robots.

Remote Triggered Labs (RTL) enable students to conduct laboratory experiments virtually. This paper demonstrates computer vision-based RTL for the Conservation of Mechanical Energy experiment. Linear regression applied to the CV-based implementation achieved optimal MSE approximately 10× better than the IR-based approach, validating CV as a scalable solution for remote laboratory access.

A system for accurately determining AQI in urban environments with high spatial and temporal variability. Includes a mobile platform with multi-camera data acquisition (front, rear, 360°), PM₂.₅/PM₁₀ sensors, noise sensors, and contextual sensors. An edge computing device processes data using deep learning pipelines and a transformer-based encoder with multi-head attention for six-category AQI classification.

Automated AQI determination using multimodal data fusion and a transformer-based deep learning architecture. Receives synchronized video streams from two cameras with environmental sensor data. Features extracted via a frozen CNN are combined with sensor readings and contextual metadata, then fed into a custom transformer for six-category AQI classification.

A remote laboratory system that uses computer vision to evaluate and grade physics experiments conducted remotely. Captures experimental setups via camera, applies CV models to determine physical outputs (velocity, trajectory), and compares against expected values — enabling scalable, sensor-free remote laboratory access.