01
What this is
A coordination layer above navigation and below human mission intent.
Research snapshot · August 2026
Lunar machines need a team operating system.
We reviewed the people and systems trying to let many different robots and spacecraft share goals, divide work, survive broken links, and continue safely without one human driving each machine.
02
Why now
Future lunar operations will mix rovers, landers, satellites, sensors, power assets, and equipment from different vendors.
03
Bottom line
The pieces exist. The complete system does not—at least publicly.
The emerging stack
Not a swarm brain. A federated hierarchy.
The strongest systems distribute routine reasoning but keep dangerous actions behind local safety rules or stricter authority.
1 · HUMAN INTENT“Survey this zone.”
2 · CAPABILITYWho can sense, move, carry, relay, or supply power?
3 · ALLOCATIONMachines bid, elect a leader, or form a temporary team.
4 · EXECUTIONEach machine acts locally and enforces its own safety limits.
5 · REPLANWork changes when links, energy, terrain, or machines fail.
6 · EVIDENCETelemetry records what happened and why.
→ The real design question: what continues safely when the network splits?
People and programs
Five strong reference points.
Federico Rossi · JPL CADRE
Three lunar rovers, elected-leader planning, distributed execution, shared state, leader transfer, and local constraint enforcement.
Flight hardware tested; lunar team operation not yet proven in the cited record.
Primary projectAli Agha · JPL NeBula
Heterogeneous wheeled, tracked, legged, and flying robots operating with no GPS and poor communications in subterranean environments.
Large physical field demonstrations.
Primary projectJonathan How · MIT ACL
Decentralized bidding, consensus, communication-aware tasking, and delay-tolerant multi-vehicle trajectory planning.
LabSimone D’Amico · Stanford SLAB
Flight-proven shared navigation using peer observations across NASA’s Starling spacecraft.
LabJulie A. Adams · Oregon State
Evidence that one operator can supervise more than 100 heterogeneous robots through mission-level commands.
ProfileWhat has actually flown
The proof is layered—not complete.
NASA Starling / DSA
Three operational CubeSats ran distributed allocation and reached consensus over a peer network. Autonomous orbital maneuvers were not completed.
FLIGHT SIGNALFlight results
ESA Proba-3
Two spacecraft repeatedly maintained precision formation using leader–wingman control, staged navigation confidence, and automatic separation safeguards.
OPERATIONAL SIGNALESA result
CAPSTONE / CAPS
Demonstrated peer-to-peer lunar navigation with LRO—but not fleet task planning or joint autonomous maneuvers.
CISLUNAR SIGNALNASA result
PRISMA / SAFE
Proved autonomous formation acquisition and reconfiguration with one active coordinator and a simpler partner.
FLIGHT SIGNALFlight results
Commercial systems
Hivemind, Nemyx, SwarmOS, and Lattice report heterogeneous teaming and field exercises. Public architecture and independent evidence remain uneven.
EARLY PATTERNOpen-RMF
A useful open interoperability baseline for structured facilities, but its dispatcher-centered design is not degraded-network lunar autonomy.
BASELINE, NOT EQUIVALENTRepository
The unresolved opportunity
No public system closes the whole loop.
The missing product is a trustworthy coordination layer that works across unlike lunar machines, survives partitions and attrition, preserves local safety authority, and leaves evidence humans can audit.
Starweave’s possible edge
Unifying heterogeneous coordination, degraded-network operation, and one-to-many human command.
What is still unproven
Scale, true decentralization, behavior during total link loss, cross-vendor integration, flight qualification, and assurance evidence.
The diligence question
Which decisions are genuinely distributed—and which still require a coordinator or Earth?
What happens next
A quiet weekly watch for real change.
We now monitor the field for new flight or field demonstrations, papers, software releases, contracts, and key personnel moves. The review separates proof from simulation and marketing.
Only report material deltas
- New hardware or flight evidence
- Architecture or safety details
- Demonstrated degraded-link behavior
- Cross-vendor integration
- Named technical leaders and maintainers
Signal test
- SIGNAL: demonstrated, documented progress
- EARLY PATTERN: repeated but incompletely verified
- NOISE: promotional language without technical substance
Monitoring lens: Lunar Outpost Starweave software.
Primary references
Source trail.
- Lunar Outpost — Software / Starweave
- JPL — CADRE planning, scheduling, execution architecture
- JPL — NeBula / CoSTAR system
- NASA — Distributed Spacecraft Autonomy flight results
- Stanford SLAB — StarFOX flight results
- ESA — Proba-3 precision formation result
- NASA — CAPSTONE/LRO navigation test
- DLR — PRISMA SAFE flight results
- Penn GRASP — MOCHA open implementation
- Open-RMF repository
Prepared as a plain-language orientation—not a claim that announced capabilities are operationally proven.