R. P. S. Hada

Research

Localisation without GPS, scheduling on constrained hardware, and learning distributed across small devices.

01Ongoing

  • HM-FGL

    Hybrid 3D localisation

    A hybrid localisation algorithm combining particle swarm optimisation with factor graph optimisation for three-dimensional wireless sensor network localisation. In development.

    PSOFactor graphs3D

  • Optimisation

    Metaheuristics for localisation

    Comparative study of metaheuristic optimisers applied to sensor node localisation, covering PSO, SSA, ACO, MVO, CMA-ES and differential evolution.

    BenchmarkingMetaheuristics

  • Indoor

    WiFi RSSI fingerprinting

    Indoor positioning from WiFi RSSI fingerprints for assisted-living settings, aimed at elderly and Alzheimer's care.

    Assisted livingRSSI

  • LoRaWAN

    MAC-layer traffic classification

    Machine learning applied to MAC-layer traffic classification in LoRaWAN networks.

    LoRaWANMAC layer

  • Federated

    Learning on edge clusters

    Federated learning across Raspberry Pi clusters for Hindi NLP and hate-speech detection, evaluating FedAvg and FedProx under non-IID data.

    FedAvgFedProxNon-IID

  • Vision

    3D indoor localisation with cameras

    Integrating camera input with radio-based estimation to improve three-dimensional indoor position accuracy.

    Sensor fusionComputer vision

  • Wildfire

    AI-assisted wildfire detection

    Continuing the Melghat Tiger Reserve line of work on early wildfire detection using sensing and machine learning.

    Environmental sensingMelghat

02Doctoral

ACM TECS 2024

Dynamic cluster head selection in wireless sensor networks

Objective. Improve energy efficiency and network lifetime in wireless sensor networks through a dynamic cluster head selection algorithm driven by residual energy, distance and node density.

Contribution. An adaptive cluster head rotation strategy that reduces energy drain over continuous sensing rounds, evaluated across large-scale simulated topologies.

Result. Extends network longevity by more than 24% against traditional LEACH protocols and mitigates premature node failure.

With Prof. Abhishek Srivastava, IIT Indore

3 figures — scroll →

Iterative localisation across deployment rounds.
Fig. 1 — Iterative localisation across deployment rounds.
Network topology and cluster formation.
Fig. 2 — Network topology and cluster formation.
Field deployment site, annotated.
Fig. 3 — Field deployment site, annotated.

Linux · Edge

Task scheduling for asymmetric multi-core edge processors

Objective. Optimise task allocation in Linux on asymmetric multi-core edge devices such as big.LITTLE architectures, matching task priority to core performance.

Contribution. A priority-aware task dispatch mechanism resolving sub-optimal Linux CFS assignment on heterogeneous cores.

Result. Up to 16% reduction in task execution time for critical real-time edge workloads.

With Prof. Abhishek Srivastava, IIT Indore

3 figures — scroll →

Cluster architecture and base station topology.
Fig. 4 — Cluster architecture and base station topology.
Scheduling behaviour across heterogeneous cores.
Fig. 5 — Scheduling behaviour across heterogeneous cores.
Prototype hardware: XBee, GPS and Arduino testbed.
Fig. 6 — Prototype hardware: XBee, GPS and Arduino testbed.

ACM TOSN 2025

A hybrid approach for localisation of sensor nodes in remote locations

Objective. Estimate node positions accurately in large-scale remote WSN deployments without GPS receivers.

Method. Loosely couples random forest machine learning with multilateration over RSSI signals, achieving both high initial accuracy and iterative multi-hop coverage.

Validation. Demonstrated on simulated datasets and prototyped on physical sensor hardware.

With Uttkarsh Aggarwal and Prof. Abhishek Srivastava

3 figures — scroll →

Localisation error across node densities.
Fig. 7 — Localisation error across node densities.
Multilateration geometry over a multi-hop network.
Fig. 8 — Multilateration geometry over a multi-hop network.
Two-phase pipeline: random forest, then iterative multilateration.
Fig. 9 — Two-phase pipeline: random forest, then iterative multilateration.