Cyber-physical IoT systems: protocol, realtime kernel and an application

Abstract

Cyber-Physical IoT (CPIoT) systems face persistent challenges in device interoperability, predictable real-time execution, and the development of accessible, context-aware applications. This dissertation addresses these challenges through three key contributions: (i) a manifest-driven application-layer IoT-device management, (ii) a memory-safe real-time kernel, and (iii) a high-precision indoor positioning system. Together, these components offer practical and scalable solutions for building resilient, safe, and inclusive CPIoT infrastructures. In Part 1, this research tackles the challenge of IoT device heterogeneity by introducing a manifest-based platform and application-layer protocol. It enables a single IoT Control App (ICA) to dynamically generate user interfaces for any compliant device, eliminating the need for multiple ICAs. By defining a manifest grammar for error-free device descriptions, the approach reduces smartphone resource drain, simplifies onboarding, and fosters seamless integration—paving the way for scalable, user-friendly consumer IoT ecosystems. In Part 2, this work presents a Rust-based hard real-time kernel for safety-critical CPIoT systems. By using compile-time safety, boolean-vector scheduling, and modular resource management, HarSaRK achieves deterministic, low-jitter task execution. Supporting single and dual-core microcontrollers with efficient IP connectivity, it outperforms conventional C-based kernels in latency and context switch times. Experimental results on STM32 and NXP platforms confirm its suitability as a safe, predictable execution foundation for nextgeneration CPIoT deployments. In Part3, a UWB-based low-cost high-precision indoor positioning system (UIPS) is introduced. This finds uses in real-time navigation in GPS-denied spaces. We implemented such IPS to support visually challenged persons navigation. Using commercial UWB modules and ToF trilateration, it achieves sub-meter accuracy with minimal anchors. A neural network classifier distinguishes LOS/NLOS conditions, improving ranging reliability. All three works have been rigorously experimented on relevant platforms, demonstrating their feasibility and practical integration into real-world CPIoT environments.