TY  - JOUR
AU  - Prakash, Satya
AU  - Nayak, Om Prakash
AU  - Patra, Om Prakash
AU  - Panigrahy, Manas Ranjan
AU  - Dass, Ashish Kumar
PY  - 2026
DA  - 2026/09/03
TI  - EcoPulse-India: An AIoT Framework for Scalable, Privacy-Compliant Urban Air Quality Monitoring on Public Transit Fleets
JO  - Next-Generation Computing Systems and Technologies
T2  - Next-Generation Computing Systems and Technologies
JF  - Next-Generation Computing Systems and Technologies
VL  - 2
IS  - 3
SP  - 89
EP  - 98
DO  - 10.62762/NGCST.2026.429196
UR  - https://www.icck.org/article/abs/NGCST.2026.429196
KW  - air quality monitoring
KW  - AIoT
KW  - edge computing
KW  - physics-informed neural networks
KW  - multi-agent reinforcement learning
KW  - data privacy
KW  - DPDPA 2023
KW  - urban sensing
KW  - India
AB  - India's stationary air-quality monitoring network remains sparse, particularly in underserved urban and peri-urban areas. EcoPulse-India addresses this limitation by using public transit buses as mobile sensing platforms within an integrated artificial intelligence of things (AIoT) framework. The framework integrates velocity-compensated aerodynamic correction (VCAC), a physics-informed convolutional graph neural network (PI-ConvGNN), multi-agent reinforcement learning (MARL)-based coverage optimisation, and privacy-respecting inference and masking architecture (PRIMA). VCAC compensates for motion-related measurement bias using vehicle kinematic and environmental information, while the quantised int8 PI-ConvGNN enables lightweight edge inference on an ESP32-S3 with a model size of approximately 38~KB. MARL prioritises underserved areas using the vulnerability-weighted coverage index (VWCI), whereas PRIMA applies on-device spatial perturbation and trajectory segmentation to reduce location-privacy risks. Simulation experiments seeded with real-world GPS traces from Delhi DTC and Mumbai BEST fleets showed that VCAC reduced mean absolute error by 22% relative to uncompensated measurements against reference-station data. MARL routing incentives raised city-level mean VWCI to 0.79---a 36.2% relative improvement over the static-routing baseline. Edge inference required 45~ms per pass with average power consumption below 0.8~W. PRIMA retained 94% of grid-cell-level spatial information required for pollution interpolation. These results demonstrate the potential of EcoPulse-India as a lightweight, privacy-aware mobile air-quality monitoring framework for resource-constrained urban environments.
SN  - 3070-3328
PB  - Institute of Central Computation and Knowledge
LA  - English
ER  - 

@article{Prakash2026EcoPulseIn,
  author = {Satya Prakash and Om Prakash Nayak and Om Prakash Patra and Manas Ranjan Panigrahy and Ashish Kumar Dass},
  title = {EcoPulse-India: An AIoT Framework for Scalable, Privacy-Compliant Urban Air Quality Monitoring on Public Transit Fleets},
  journal = {Next-Generation Computing Systems and Technologies},
  year = {2026},
  volume = {2},
  number = {3},
  pages = {89-98},
  doi = {10.62762/NGCST.2026.429196},
  url = {https://www.icck.org/article/abs/NGCST.2026.429196},
  abstract = {India's stationary air-quality monitoring network remains sparse, particularly in underserved urban and peri-urban areas. EcoPulse-India addresses this limitation by using public transit buses as mobile sensing platforms within an integrated artificial intelligence of things (AIoT) framework. The framework integrates velocity-compensated aerodynamic correction (VCAC), a physics-informed convolutional graph neural network (PI-ConvGNN), multi-agent reinforcement learning (MARL)-based coverage optimisation, and privacy-respecting inference and masking architecture (PRIMA). VCAC compensates for motion-related measurement bias using vehicle kinematic and environmental information, while the quantised int8 PI-ConvGNN enables lightweight edge inference on an ESP32-S3 with a model size of approximately 38~KB. MARL prioritises underserved areas using the vulnerability-weighted coverage index (VWCI), whereas PRIMA applies on-device spatial perturbation and trajectory segmentation to reduce location-privacy risks. Simulation experiments seeded with real-world GPS traces from Delhi DTC and Mumbai BEST fleets showed that VCAC reduced mean absolute error by 22\% relative to uncompensated measurements against reference-station data. MARL routing incentives raised city-level mean VWCI to 0.79---a 36.2\% relative improvement over the static-routing baseline. Edge inference required 45~ms per pass with average power consumption below 0.8~W. PRIMA retained 94\% of grid-cell-level spatial information required for pollution interpolation. These results demonstrate the potential of EcoPulse-India as a lightweight, privacy-aware mobile air-quality monitoring framework for resource-constrained urban environments.},
  keywords = {air quality monitoring, AIoT, edge computing, physics-informed neural networks, multi-agent reinforcement learning, data privacy, DPDPA 2023, urban sensing, India},
  issn = {3070-3328},
  publisher = {Institute of Central Computation and Knowledge}
}