Autonomous mobile robots increasingly rely on deep learning models and distributed computing infrastructures spanning edge, fog, and cloud layers. In healthcare environments, these systems must meet strict constraints regarding latency, power consumption, and operational management, while remaining safe and robust. This thesis investigates the implementation of Artificial Intelligence (AI)-based autonomous mobile robots on heterogeneous multi-layer platforms by jointly addressing the on-board execution of deep learning models, real-time data distribution, and the dynamic orchestration of distributed robotic services. The work relies on a realistic testbed, combining a modernized mobile robot and three Kubernetes clusters representing the different distributed layers. A deep learning-based visual odometry model is first optimized for real-time execution, demonstrating that such performance can be achieved without degrading localization accuracy, while quantifying the trade-offs in terms of resources, energy, and temperature. A multi-layer communication middleware, BlazeFlow, is then introduced to unify data flows between heterogeneous publish-subscribe protocols and ensure low-latency communication for high-throughput sensor streams between the edge, the fog, and the cloud. Building on this, BlazeAIoT is proposed as a modular, multi-layer platform integrating communication, configuration, monitoring, and deployment of robotic services, and validated through navigation and emergency stop scenarios, where services are distributed across fog and cloud edge clusters while maintaining real-time performance. Finally, BlazeMaestro is presented as a digital twin that combines a simulator with a multi-cluster deployment to evaluate orchestration strategies that account for resource, latency, and energy constraints, and to compare metaheuristic, deep reinforcement learning, greedy, and autoencoder-based orchestrators. Overall, this research provides tools and experimental evidence to guide the deployment of autonomous mobile robots on heterogeneous multi-layer infrastructures and clarifies how to balance latency, energy, and operational complexity in smart healthcare environments.
Melançon, C. (Author),
Saad (Supervisor),
Kaur (Co-supervisor) &
Gascon-Samson (Co-supervisor),
6 Aug 2026Student thesis: Doctoral thesis › Doctorate in Engineering: Engineering