Knowledge Graph
Core Team
Global Partners / Competitors
Research Pillars
Institutions / Funding
Executive Summary
A comprehensive global landscape analysis confirms a clear white space: no research group worldwide frames space systems as autonomous organizations. While multi-agent coordination, LLM agents for space, formal verification, and satellite swarm autonomy are all active areas individually, no program combines organizational MAS theory, agentic AI/LLM planning, formal safety verification, and heterogeneous physical space asset coordination into a unified architecture.
This four-dimensional gap — confirmed across Germany, Europe, the US, and the rest of the world — is the unique space AOIS proposes to occupy. The DFG funding moment is optimal: SPP 1914 (Cyber-Physical Networking) closed in 2023, and no Forschergruppe, SFB, or SPP currently addresses multi-agent autonomous space systems.
Strategic path confirmed: FOR (Forschergruppe) as the first concrete step — sketch to be prepared summer 2026. Closest competitors are JPL/CADRE (homogeneous rovers only), ONERA/Picard (MAS for Earth observation), and Zhejiang Lab (distributed computing, no organizational theory). None address the full AOIS intersection.
Next meeting: Thu 19 March, 12:30–15:30, Alessandro's office.
The Big Question (Confirmed)
✅ Final Framing (12 Mar 2026)
"Can space systems be engineered as autonomous multi-agent architectures that negotiate, self-organize, and coordinate heterogeneous physical assets — satellites, orbital robots, rovers, ground stations, life support — without centralized human orchestration?"
Decision log:
✗Option A: Multi-agent systems in space (too standard)
✗Option B: Provably safe AI with space as testbed (too AI-centric)
△Option C: Engineering autonomy under physical constraints (correct direction but not distinctive enough)
✓Option D (confirmed): Organizational intelligence — not multi-agent planning. The term "organization" (roles, delegation, negotiation, governance) sets AOIS apart from all identified competitors worldwide.
Key differentiator: AOIS is not "multi-agent planning" (crowded: JPL, ONERA, CMU) and not "swarm intelligence" (crowded: ESA ACT, NASA Ames). AOIS introduces organizational intelligence as a formal engineering discipline for space — roles, delegation, self-governance as first-class concepts. No space group does this.
Frame as: "CADRE for heterogeneous assets" — NASA/JPL's CADRE deploys 3 homogeneous rovers. AOIS extends this to fundamentally different asset types (satellites, orbital robots, rovers, habitats, propulsion) with richer organizational autonomy and formal guarantees.
Landscape Intelligence
The 4-Dimensional Gap (confirmed)
No group worldwide combines all four: (a) Organizational MAS theory applied to space; (b) Agentic AI/LLM multi-agent coordination (all existing work is single-agent); (c) Formal safety verification of multi-agent negotiation and delegation; (d) Physical asset heterogeneity — coordination across satellites, robots, rovers, habitats as peers.
Closest competitors
JPL CADRE / FAME — 3 homogeneous rovers + federated scheduling across satellites. World's most advanced multi-agent space demo. Gap: homogeneous agents only; no organizational theory; no LLM reasoning layer.
ONERA/Picard — multi-agent federation for Earth observation, MASSpace co-organizer. Gap: Earth observation scheduling only; no physical heterogeneity; no formal verification.
Zhejiang Lab (China) — 12 satellites, 10 AI models on-orbit (Three-Body, launched May 2025). Gap: distributed computing focus; no organizational theory, no negotiation, no heterogeneous assets.
Differentiation strategy
Use "space systems as autonomous organizations" — negotiation, delegation, and self-governance as first-class engineering concepts. This framing is novel and immediately communicates AOIS's distinction.
Avoid: "swarm intelligence" (crowded), "constellation management" (too narrow), "autonomous operations" (single-agent connotation).
Emphasize holonic MAS (HMAS) as theoretical basis — every agent is simultaneously part and whole. No space group has applied holonic architecture to physical space systems. Recent Ashfaq et al. (2025) LLM+holonic SoS paper is the closest conceptual predecessor.
Global Network
Priority Collaborators
- Collab ONERA / G. Picard MASSpace co-chair · closest European work · AAMAS 2024
- Collab TU Delft / E. Gill MAS onboard autonomy · AltiCube+ · 300+ publications
- Collab PoliMi / M. Lavagna ERC EXTREMA · ESA multi-agent contracts · DART Lab
- Collab JPL / S. Chien · F. Rossi CADRE · FAME · CASPER · MASSpace chair
- Collab Stanford / M. Kochenderfer Decision under uncertainty · aerospace AI
- Collab Tohoku SRL / K. Yoshida MoonBot Moonshot · DLR LUNA collab
- Collab TU Berlin / W. Hönig Emmy Noether · heterogeneous swarm coord.
- Collab Technion / P. Gurfil SAMSON · world-first autonomous nanosatellite formation
Likely DFG Peer Reviewers
- Reviewer Steve Chien JPL Fellow · world's leading AI-for-space expert
- Reviewer Kristin Rozier Iowa State · formal verification · R2U2
- Reviewer Gauthier Picard ONERA · MAS for space · MASSpace co-founder
- Reviewer Eberhard Gill TU Delft · distributed space systems
Competitors
- Compete Zhejiang Lab 3-Body · 12 sat · 10 AI models · May 2025
- Compete DARPA Blackjack / Pit Boss Constellation autonomy manager
- Compete MDA / Canadarm3 AI-driven autonomous Gateway robotics
Funding Pipeline Strategy
Summer 2026 → Submit
FOR — Forschergruppe ★ First Step
Sketch summer 2026. Research Unit as fastest path to DFG funding. ~0.5–2M€/yr. Confirmed gap: no FOR, SFB, or SPP currently addresses multi-agent autonomous space systems. Optimal moment.
→
2027–2028
SFB Sketch
Sonderforschungsbereich — 2–3M€/yr, 4+4+4 years. FOR track record feeds into SFB. No round (continuous submission). First SFB ever led from TUM-ASG.
→
2027+
Horizon Europe + ERC
EU Cluster 4 (Space) calls. ERC Synergy Grant (up to 10M€). Parallel track — builds European network. ONERA/Picard, TU Delft/Gill, PoliMi/Lavagna as natural partners.
→
~2029–2030
Cluster of Excellence
Round 3 sketch. Arrive with FOR/SFB running, publications, proven consortium. Target funding ~2033.
DFG gap confirmed: 8 new Research Units established Dec 2024, 9 more Dec 2025 — none on space multi-agent systems. 8 new SPPs March 2025 — none relevant. SPP 1914 (Cyber-Physical Networking) closed 2023. Window is open now.
Core Team
- PI Alessandro Golkar Systems architecture, federated systems, AI agents → Spokesperson
- PI Marcello Romano Orbital robotics, GNC, proximity ops → Physical agent actions
- Team Gisela Detrell Life support (ECLSS) → Bio-agent integration
- Team Chiara Manfletti Propulsion, space traffic mgmt → Agent traffic governance
- Team Philipp Reiss ISRU, exploration → Shared resource coordination
- Team Niklas Kochdumper Formal verification, AI safety → Verified agent organizations
- Co-PI? Matthias Althoff TUM CPS — formal safety, reachability → Complementary verification layer
- Co-PI? Felix Brandt TUM Decision Science — mechanism design, voting/allocation → Agent negotiation theory
⚠️ None currently involved in active SFB or Cluster. No SFB has ever been led from TUM-ASG. Althoff and Brandt strengthen the FOR's theoretical depth and broaden DFG appeal.
Open Points
- Critical Prepare FOR sketch — target summer 2026 submission. Assign lead writer per research area.
- Critical Confirm Althoff and Brandt as Co-PI candidates — approach informally before Thu meeting
- Open Contact ONERA/Picard and TU Delft/Gill for collaboration interest letter
- Open Define 4–5 research areas with PI assignments for FOR structure
- Open Each PI: 3–5 bullets on unique contribution to AOIS organizational intelligence thesis
- Open Submit abstract / position paper to MASSpace'26 (May 2026, Paphos) — Golkar on Programme Committee
- Done Big question confirmed: organizational intelligence framing
- Done Instrument confirmed: FOR as first step (sketch summer 2026)
- Done Global landscape analysis completed — white space confirmed
- Done Closest competitors identified (JPL CADRE/FAME, ONERA/Picard, Zhejiang Lab)
- Done Differentiation strategy: "organizational intelligence" not "multi-agent planning"
- Done DFG funding gap confirmed — no active FOR/SFB/SPP on topic
Actions for Thu 19 March Meeting
🎯 Priority 1: Lock FOR structure — define 4–5 research areas, assign PIs, confirm Althoff and Brandt as Co-PI candidates. The FOR sketch needs to start this summer.
📋 Priority 2: Validate "organizational intelligence" framing with full team. Does it resonate with all six PIs? Is "CADRE for heterogeneous assets" a useful communication anchor?
🌍 Priority 3: Identify 2–3 international co-applicants or associated partners — ONERA/Picard and TU Delft/Gill are highest priority. Approach for informal interest before FOR submission.
📝 Before meeting: Each PI writes 3–5 bullets on their unique contribution to the AOIS thesis. This shapes the FOR's research area structure.
🗓️ MASSpace'26: Submit position paper to MASSpace'26 (May 26, Paphos, Cyprus) — Golkar on Programme Committee. First visibility for AOIS in the international community.
Strategic Insights
💡 The "organizational" angle is timing-perfect: LLM agents for space exploded in 2024–2025 (ASTREA on ISS, JPL+Anthropic on Mars Perseverance), but ALL demonstrations are single-agent. Multi-agent LLM coordination across heterogeneous space assets — the AOIS core — is unclaimed territory.
💡 Holonic MAS as theoretical anchor: HMAS (holonic multi-agent systems) — every agent is simultaneously part and whole — has never been applied to physical space systems. Ashfaq et al. (2025) LLM+holonic SoS paper is the conceptual predecessor. AOIS is the space-domain instantiation.
💡 Kochdumper + Althoff = uniquely strong safety case: Formal verification of multi-agent organizational decisions is an open problem. Two TUM chairs on reachability analysis and formal methods makes AOIS's safety layer uniquely credible in DFG review.
💡 Villa Vigoni network as industry board: The Sep 2026 workshop participants (Isar Aerospace, D-Orbit, OHB, OroraTech, Alpine Space Ventures) form a natural industry advisory board for the FOR/SFB.
💡 MASSpace'26 is the right first venue: Workshop chairs are Steve Chien (JPL) and Gauthier Picard (ONERA) — the two most likely DFG peer reviewers. Presenting at MASSpace'26 establishes AOIS visibility directly with the reviewers who matter.