My research sits at the intersection of safety-critical artificial intelligence, physics-based modelling, sensing, biomechanics, and functional safety.

Two connected research streams currently define much of my work:

I investigate how vehicle safety systems can better detect, interpret, and respond to difficult or atypical safety scenarios, particularly non-upright pedestrians and vulnerable occupants.

This programme connects:

• multimodal sensing and sensor fusion;
• computer vision and uncertainty-aware AI;
• forensic biomechanics;
• pedestrian-impact reconstruction;
• injury-risk modelling;
• autonomous vehicle and ADAS safety; and
• ISO 26262-aligned functional-safety architectures.

The broader objective is to move from post-event forensic reconstruction toward proactive detection, intervention, and injury prevention.

A second major research stream examines battery State-of-Health (SoH) from both machine-learning and functional-safety perspectives.

My recent work investigates:

• battery SoH estimation;
• validation inflation and model generalisation;
• cross-cell transfer reliability;
• predictor structure;
• battery-health uncertainty;
• EV power-margin integrity; and
• the role of battery condition in safety-critical vehicle functions.

The central question is not only whether a model performs accurately on familiar data, but whether its predictions remain reliable when transferred to previously unseen cells and real operating conditions.

Research principle: High apparent accuracy is not enough for a safety-critical system. The system must remain reliable when the operating conditions, data distribution, or physical system changes.

Earlier and parallel research includes geospatial intelligence, remote sensing, satellite-data analysis, hydrological forecasting, computational astrophysics, stellar-population modelling, and galactic evolution.

Sensors · DOI · Open Access

NUP-REPORT 1.0: A Proposed Reporting and Benchmarking Framework for Non-Upright Pedestrian Detection and Pre-Crash Safety Evaluation

NUP-REPORT 1.0 proposes a structured reporting and benchmarking framework for non-upright pedestrian detection and pre-crash safety evaluation.

The framework addresses a fundamental methodological challenge in safety-critical perception research: apparently similar detection results can be difficult to compare when studies differ in posture definitions, scenario coverage, sensor and data provenance, timing conventions, uncertainty reporting, and vehicle-response assumptions.

NUP-REPORT 1.0 introduces six reporting domains, a 30-item checklist, a scenario-coverage matrix, an explicit five-timestamp event model, detection-referenced stopping formulations, and a 20-publication feasibility audit.

Its central principle is that detection accuracy is not equivalent to safety performance. The framework is intended to support more transparent reporting, reproducible evaluation, meaningful cross-study comparison, and future development of shared benchmarks for difficult-to-detect vulnerable road users.

Scientific Reports · DOI · Open Access

Advanced machine learning techniques for daily streamflow forecasting: a case study of the Brahmaputra River

This study investigates machine-learning-based daily streamflow forecasting for the Brahmaputra River, comparing multiple modelling approaches across forecast horizons of up to 30 days.

The work focuses not only on predictive performance, but also on lead-time-dependent behaviour, temporal robustness, uncertainty, and the practical limitations of data-driven river-flow forecasting.

Scientific Reports · Open Access

Battery State-of-Health as a Functional Safety Variable: an ISO 26262-Aligned AI Framework for Electric Vehicle ADAS Power Integrity

This research treats battery State-of-Health as a functional-safety variable, rather than solely as a maintenance indicator.

It presents a five-layer architecture linking:

SoH estimation → power-margin monitoring → safety decision logic → vehicle response → lifecycle management

The framework examines how battery degradation and auxiliary electrical demand may affect the power margin available to safety-critical ADAS functions.

A Physics-Grounded Multi-Modal Sensor Fusion Framework for Pedestrian Impact Kinematic Reconstruction Under Uncertainty: Phase 1 Design and Theoretical Evaluation

This work develops a physics-grounded framework combining multimodal sensing, kinematic reconstruction, uncertainty propagation, and forensic interpretation for pedestrian-impact analysis.

A Multi-Modal AI System for Detecting Pedestrians Lying on the Road: Simulation-Based Safety and Injury Risk Analysis

This research investigates one of the difficult edge cases in automated vehicle safety: detecting pedestrians who are already lying on the road.

The work combines multimodal sensing, AI-based detection, simulation, and injury-risk analysis to examine how improved perception could support earlier intervention.

• Autonomous vehicle and ADAS safety
• Detection of fallen and non-upright pedestrians
• Vulnerable-road-user protection
• Safety-critical perception
• Functional-safety decision systems
• ISO 26262-aligned safety architectures

• Multimodal artificial intelligence
• Sensor fusion
• Computer vision
• Machine learning
• Uncertainty-aware decision systems
• Model validation and generalisation
• Cross-domain and cross-system transfer

• Pedestrian-impact reconstruction
• Injury mechanisms
• Vehicle-occupant safety
• Wheelchair occupant protection
• Kinematic reconstruction
• Uncertainty quantification
• Accident reconstruction

• Battery State-of-Health estimation
• Battery-health prediction
• Validation inflation
• Predictor structure
• Cell-to-cell transfer reliability
• EV power-integrity modelling
• Battery-informed functional safety

• Geospatial intelligence
• Remote sensing
• Satellite-data analysis
• Hydrological forecasting
• Environmental time-series modelling
• Computational astrophysics
• Stellar-population modelling
• Galactic evolution

View on Google Scholar

• From Post-Mortem to Prevention: Redefining “Invisible” Pedestrians through ISO 26262 and Multi-Modal AI — SSRN, 2026
• Integrated Safety Architectures: Leveraging Multi-Modal AI and ISO 26262 to Protect Vulnerable Road Users — SSRN, 2026
• Sudden Incapacitation or Death at the Wheel: Unravelling the Predictors of Catastrophic Multi-Vehicle Collisions — SSRN, 2026

• Mechanisms, Modelling, and Machine Learning-Based Prediction of Lithium-Ion Battery Degradation in Electric Vehicles: A Comprehensive Review — SSRN, 2025

• Global Homeland Security Satellite Imagery Market: Strategic Outlook and Growth Trajectories — SSRN, 2025
• SATCOM, the Future UAV Communication Link — SSRN, 2022

• Galactic Archaeology: A Chemo-Kinematic Review of the Milky Way's Hierarchical Assembly — SSRN, 2025
• Galactic Paleontology: Reconstructing Accretion Events with Chemo-Dynamical Signatures — SSRN, 2025
• Unveiling Galactic Assembly: Chemo-Kinematic Insights from Stellar Absorptions — SSRN, 2025

I welcome research and technical collaboration in:

• safety-critical artificial intelligence;
• automotive and autonomous-system safety;
• vulnerable-road-user protection;
• multimodal sensing and sensor fusion;
• forensic biomechanics and accident reconstruction;
• ISO 26262 and functional-safety engineering;
• electric-vehicle battery intelligence;
• battery State-of-Health estimation and validation;
• geospatial intelligence and remote sensing;
• hydrological and environmental machine learning;
• scientific computing; and
• computational astrophysics.

For research, technical, or professional enquiries, please connect with me through:

LinkedIn

“From forensic reconstruction to proactive prevention — and from model accuracy to trustworthy real-world performance.”

Dr. Nick Barua

Google Scholar · ORCID · LinkedIn