FAQ
What does Arboria Labs work on?
Decentralized coordination under real communication constraints — propagation delay, finite bandwidth, energy per bit, and packet loss — at agent counts where a central coordinator is not an option. Our measured results establish where coordination collapses as delay grows (Phase Diagram of Coordination Under Delay) and how cheap peer-state prediction pushes that boundary outward (Anticipatory Coordination via Peer-State Prediction). We apply the result to orbital compute constellations, where the delay that matters is not light-lag but contact-wait — the time a message spends waiting for an inter-satellite link to exist at all. See the research index.
Is the tooling open source?
No. The Leviathan Engine, Gossamer Threaded Intelligence, and Maneuver.Map are proprietary and are not distributed. Their documentation pages describe the interfaces, configuration surfaces, and models they implement in enough detail to understand and evaluate our results, and to reimplement the experiments against another simulator. The algorithms themselves are drawn from the open literature and are covered in Techniques and Foundations.
How large a swarm can Leviathan simulate?
Both of the engine’s former quadratic hotspots — neighbour-finding for the communication model and for collision resolution — are now resolved through a spatial grid, so per-step cost scales linearly in agent count. In practice the ceiling is memory and patience rather than algorithmic complexity. Our published coordination results deliberately run at small N (500 and 2,000): the point of that work was to show the delay-collapse boundary is invariant in swarm size, which is a claim you make by varying N, not by maximizing it.
How do I reproduce a paper?
Each paper’s reproducibility appendix names the canonical batch, the full experiment grid, the fixed parameters, and the pinned environment — package versions, wheel digest, base-image digest, and the lockfile hash that fingerprints the worker. Because the engines are proprietary, you cannot rerun them yourself. We share the experiment specifications and result records with reviewers and collaborators on request. The full policy, including what is deliberately not in the provenance block, is at Reproducibility and Data Availability.
Do you collaborate with universities?
Yes — we’re actively open to student collaborations, joint papers, and shared benchmarks. Reach out at community@arborialabs.com and see the Community page for labs and testbeds we work with.
What counts as “dual-use” research at Arboria?
Swarm methods are inherently dual-use. We do not contribute directly to lethal autonomous weapons systems and we gate publications that would materially accelerate them. For planetary and orbital work we treat Kessler-syndrome risk and forward contamination as first-order design constraints. See our principles.
Can I cite your work?
Please do, with one caution. Nothing is on arXiv or peer-reviewed yet; we distribute papers as Arboria Labs technical reports. Five carry measured results derived from committed batches — the four delay-coordination papers and the hierarchical-market paper. The two archive papers (ICCD and Density-Modulated Boids) document implemented mechanisms but their result tables are illustrative rather than measured, and are marked as such: cite those two for their mechanisms, not their numbers. The publications page is the canonical structured index and carries BibTeX for each.
How is Arboria different from other swarm research groups?
We model the constraint that most work assumes away. The standard setup in swarm robotics and multi-agent reinforcement learning gives each agent an instantaneous view of its neighbours; ours does not, and the difference turns out to be a phase boundary rather than a performance penalty. That framing — plus scale, decentralization, and instrumentation — is what carries into the orbital work, where a message waits minutes for a link to open even though light would have crossed the constellation in microseconds.
How do I stay updated?
Latest updates carries releases and results. For anything else, contact us.