Clemente J. Juan Oliver

About

I am a Research Associate and Doctoral Candidate in Prof. Alessandro Golkar's Chair of Spacecraft Systems at the Technical University of Munich (TUM). Since 2024 my doctoral research has focused on autonomous satellite operations with artificial intelligence: multi-agent reinforcement learning, and cognitive architectures built on large language models that reason, plan and call tools to manage large satellite constellations. I am also exploring joint-embedding predictive architectures (JEPA) as world models for space systems.

Before my PhD I earned a Bachelor's in Aerospace Engineering at the Universitat Politècnica de València (Spain) and a Master's in Aerospace Engineering at TUM, where I also worked as a Research Assistant at the University of the Bundeswehr Munich on distributed real-time hybrid systems. Outside the lab I build AI products as a founder (see Ventures).

Projects

• PhD · AUTOPS

  AUTOPS framework

  A controlled testbed for comparing satellite operations architectures along three axes: who decides (single agent, centralised, decentralised or hybrid multi-agent), how (rules, reinforcement learning, LLM agents with safety shields, calibrated decision models or learned world-model planning) and where (ground, onboard or hybrid).

• Space mission

  EventSat

  A 6U CubeSat from the Chair of Spacecraft Systems that flies a neuromorphic event camera to detect objects in orbit. I lead the on-board computer and command & data handling (OBC & CDH) subsystem.

  Mission page ›
• PhD · First-author paper

  Multi-agent RL for constellations

  Decentralised decision-making for space situational awareness under partial observability, trained with multi-agent PPO (EUCASS 2025). Within AUTOPS, these RL policies keep being improved and are benchmarked against rule-based, LLM-agent, decision-model and world-model approaches.

  Project page ›
• PhD · AUTOPS

  Onboard world models

  Latent world models in the spirit of JEPA, learned from operations data and used for onboard planning, evaluated against an exact-dynamics reference planner.

• PhD · AUTOPS

  Calibrated decision models

  Can a small, fast model choose what a satellite does next, and know when it is unsure? I adapt calibrated decision models to onboard operations, training them on simulated mission outcomes so their confidence can gate when the spacecraft acts on its own.

Publications

Conference papers

1. R. M. García Alarcia, S. Dolan, J. Sindermann, C. J. Juan Oliver, L. Schuberth, V. Messina, F. Salmaso, A. Golkar. EventSat: A CubeSat Mission for Neuromorphic Vision-Based Space Object Detection. 5th Symposium on Space Educational Activities (SSEA), 2026.
2. C. J. Juan Oliver, V. Messina, S. Dolan, A. Golkar. Autonomous Decision-Making for Large Satellite Constellations: a Multi-Agent Reinforcement Learning Approach to Space Situational Awareness in Partially Observable Dynamic Environments. 11th European Conference for Aeronautics and Aerospace Sciences (EUCASS), Rome, 2025. [PDF]
3. S. Dolan, L. Schuberth, R. Arge, R. M. García Alarcia, V. Messina, C. J. Juan Oliver, F. Salmaso, J. Sindermann, F. Sofio Avogadro, A. Golkar. Design and Analysis of an Event Camera Payload for Space-Based Object Detection on the EventSat 6U CubeSat Mission. 15th IAA Symposium on Small Satellites for Earth System Observation, Berlin, 2025.

Theses

1. C. J. Juan Oliver. Deep Reinforcement Learning for Decentralized Autonomous Decision-Making in Federated Satellite Systems. Master's thesis, TUM, 2024.
2. C. J. Juan Oliver. Evaluating the Efficiency and Scalability of CubeSat Simulations on Low-Capacity Systems: A Comparative Study. Master's term paper, TUM, 2024.
3. C. J. Juan Oliver. On Horizontal Axis Wind Turbines Design, Optimization and CFD Analysis. Bachelor's thesis, UPV, 2020.

Full and up-to-date list on Google Scholar.

Teaching

Interested in a thesis on AI for spacecraft autonomy? See the Chair's open thesis topics or email me.

Ventures

• Exploitiv

  Co-founder and lead developer. A poker learning platform for players and schools, with Héctor, an AI coach that analyses hands and explains decisions.

• Sereno

  Co-founder and lead developer. Autonomous workflows for companies: unifying business data and recommending, then executing, actions with human approval. Launching soon.