Operational AIoT: lifecycle management and responsible deployment of TinyML models in heterogeneous IoT infrastructures

Funding institution: Becas Leonardo (Investigadores y Creadores Culturales) – Fundación BBVA
Duration: 1 October 2024 – 31 March 2026

CoAIoT focuses on enabling practical, responsible and maintainable AI on IoT by providing an operational framework to deploy, monitor and evolve TinyML/edge models in heterogeneous environments (edge–fog), with special attention to model lifecycle, data drift, energy constraints, and trustworthy operation in real deployments. The project aims to reduce the gap between research prototypes and real AIoT roll-outs by offering tools, guidelines and validation assets that make on-device intelligence sustainable over time.

The design is based on three basic pillars:

– TinyML deployment at scale (edge–fog orchestration). CoAIoT adopts an edge–fog architecture where decisions can be executed close to the data source when latency matters, while fog nodes coordinate configuration, updates and resource-aware execution. The emphasis is not only “running inference”, but doing it repeatably and safely across fleets of devices.

– Model lifecycle and adaptability (drift-aware AIoT). Instead of prioritising the creation of many models as a goal in itself, CoAIoT targets the full lifecycle: selection, deployment, monitoring, feedback and controlled evolution of models as contexts change (“dynamic environments / CACE”). The platform will incorporate mechanisms to detect performance degradation, trigger corrective actions (e.g., recalibration, fine-tuning, or model replacement) and keep models aligned with their operational context.

– Observability, evaluation and responsible operation. CoAIoT will deliver an evaluation framework with KPIs and benchmarks (latency, energy, reliability, and robustness), and will validate the approach through realistic pilots (e.g., urban sensing, indoor scenarios, mobility/logistics), producing reusable assets and best practices that support real adoption beyond a single deployment.

IP and technical contact

Sergi Trilles (strilles@uji.es)

Funding