Decentralised Coordination and Task Allocation
Published:
Role: Doctoral research and thesis supervision · Chair of Spacecraft Systems, Technical University of Munich
A distributed satellite system only becomes more than the sum of its spacecraft if the spacecraft can decide among themselves which of them serves a given request. This line of work treats that decision as an optimisation over a time-varying graph: the network topology changes continuously with the orbits, so the coordination scheme, not just the constellation geometry, determines how quickly a task reaches the satellite best placed to execute it.
The recurring question is how much centralisation a network actually needs. Fully centralised coordination is responsive but brittle and expensive in downlink; fully decentralised coordination scales but converges slowly. A small fraction of central nodes turns out to recover most of the responsiveness of a centralised scheme while keeping the scalability of a decentralised one.
Related publications: Initial Formulation of a Time Varying Dynamic Graph Decentralized Optimization Framework (IAC 2023) · Latency Optimization in Centralized and Decentralized Coordination of Time-Varying Evolutionary Satellite Networks (IAC 2024) · Efficient and Responsive Task Allocation in Distributed Satellite Systems: The Role of Central Nodes (IWPSS 2025) · Advancing Federated Satellite Systems Performance: A Collaborative Method for Improved Object Detection in Space (AIAA SciTech 2025) · The Role of Central Nodes in Multi-Task Allocation (MASSpace, AAMAS 2026)
Supervised theses
Multiple Task Allocation for Federated Satellite Systems
Extended the reward-based selective propagation framework from single-task to multi-task allocation in federated satellite systems. More than 250 simulations, with up to 5,000 satellites and 500 concurrent tasks, showed convergence below 100 s with as few as 5% central nodes. The work also found a power-law relation between network size, central node fraction and link efficiency.
The Role of Central Nodes in Multi-Task Allocation for Distributed Space-Based Observation Systems
Effect of Ground Stations in Satellite Networks for Task Distribution in Centralized and Decentralized Approaches
Studied how ground stations affect task distribution in centralised, partially centralised and decentralised networks. After benchmarking 44 optimisation algorithms for constellation coverage, he found that denser ground station networks cut delays sharply in centralised setups but helped much less in decentralised ones.
Deep Reinforcement Learning for Decentralized Autonomous Decision-Making in Federated Satellite Systems: Enhancing Space Operations with AI
Developed a modular FSS simulation environment covering coordination models from fully centralised to fully decentralised. In it he trained DQN, SAC and PPO agents to observe targets, share data and manage energy and storage, in a scenario with 20 observers and 20 targets. He then benchmarked the agents on an NVIDIA Jetson to show they could run on board.
Schedule Optimization for a Heterogeneous Earth Observation Constellation
Built a scheduling optimisation for OroraTech's wildfire-monitoring constellation, starting from the FOREST-1 operational approach and preparing for FOREST-2. Scenarios ranged from two satellites to 18 orbital planes. Data budget turned out to be the limiting resource, and constellations sized plane by plane for continuous coverage fulfilled imaging orders far better than randomly assembled ones.
Schedule Optimization for a Heterogeneous Earth Observation Satellite Constellation, IAC 2023
Advancing Satellite Network Performance: Network Analysis for Federated Satellite Systems
Developed an HLA-based, hardware-in-the-loop-capable simulator for a 100-CubeSat federation with UHF inter-satellite links. The work measured how onboard resource limits affect message spreading and used graph-theoretic centrality metrics to predict it. It then tested sequential, parallel and reverse bidding in a flood rapid-mapping case (Emilia-Romagna, 2023), reaching consensus on an imager in under 8 hours.
Advancing Satellite Network Performance, IEEE Access, 2024 · Full project page