AI-driven multi-agent orchestration for distributed cloud-edge applications

Emerging digital experiences increasingly rely on distributed computation, communication and AI-driven processing across heterogeneous cloud, edge and device infrastructures. Future immersive formats, including real-time telepresence, extended reality (XR) and Gaussian Splat-based 3D media delivery, as well as industrial automation and remote healthcare applications, require end-to-end adaptation of application behaviour and infrastructure resources to satisfy stringent requirements on latency, scalability, resource efficiency and Quality of Experience (QoE). These services consist of interconnected AI models, media processing pipelines and microservices that operate across multiple domains and administrative boundaries, creating unprecedented orchestration complexity.

The objective of this PhD is to develop AI-native orchestration frameworks that enable distributed applications to autonomously optimize their full lifecycle. In particular, the research will investigate how Agentic Systems, composed of collaborative AI agents, can continuously observe system conditions, reason about trade-offs and coordinate decisions across application, cloud and network layers. Such coordination is required to jointly optimize application-level objectives, including media adaptation, rendering and AI inference, together with infrastructure-level objectives such as resource allocation, reliability, cost and energy efficiency.

A central research challenge is the design of hierarchical multi-agent architectures capable of managing large-scale, heterogeneous and geographically distributed infrastructures. The research will explore reinforcement learning (RL), multi-agent RL and hierarchical control mechanisms that enable both local autonomy and global coordination across independently managed domains. Digital twins will be leveraged to create realistic and predictive environments for training, validating and continuously optimizing orchestration policies before deployment in operational systems.

Many aspects of these agentic orchestration platforms remain largely unexplored, including coordination mechanisms, reward engineering strategies, knowledge sharing and generalization across applications and operational contexts. The expected outcome is a new generation of scalable AI-driven orchestration solutions that advance the state of the art in cloud-edge service management while enabling next-generation immersive and real-time distributed applications in sectors such as media, communications, healthcare and industry.