The local rule
Craft broadcast information about their motion, anticipate encounters and steer to keep distance. The model combines avoidance with goal following and, in selected configurations, alignment, corridors, authority rules and vehicle response limits.
The research varies these ingredients to find both improvements and failure mechanisms. A measured improvement belongs to the configuration tested; it does not automatically transfer to a different craft or environment.
How a research flight works
- Ask. State a bounded question and the competing explanations.
- Plan. Fix settings, comparisons, predictions and decision rules before reading the result.
- Fly. Run the specified seeds and preserve the outputs.
- Challenge. Review the method, arithmetic, interpretation and realism.
- Record. Integrate the result, its conditions, corrections and remaining questions.
What the model assumes
Vehicle classes are parameterised response models. Communication tiers, navigation-error processes, rigidity thresholds and many corridor dimensions are assumed. Buildings come from city datasets, with simplified solids and terrain. The simulator uses a shared clock; the proposed independent-agent architecture is future work.
Contacts, wall touches, damage, losses and completed tasks are different measurements. Their exact definitions depend on the active experiment and its configuration. A count of zero in a finite run does not establish zero risk.
The Manhattan Swarm instrument
The live instrument uses its own logistics planner and a standalone port of a specific Equilibrium physics configuration. The port assumes perfect information and omits the research link, weather, city-solid contact and mortality paths; the page supplies its own city behaviour. It is useful for exploration, but its live results are not the CITY or LINK research campaigns.
Reproduce a result
Start from a flight-log entry and its pinned source record. Use the experiment’s own settings, seeds and checks. Record the repository revision, local changes, execution environment and result files before comparing the result.
git clone https://github.com/resonant4/equilibrium.git
cd equilibrium
git checkout 41cfa930241774c0983f2850782cf541b588a23eThe repository contains both Python and JavaScript tools. Individual experiments specify their own data and commands; the main README describes the project setup.
Repository setup and experiments ↗Geometry and attribution
The city pipeline uses NYC Open Data Building Footprints (5zhs-2jue) and the Citywide Street Centerline dataset (inkn-q76z). The instrument retains its embedded source metadata. The three-dimensional renderer uses the repository’s vendored Three.js distribution.
All quantities on this website are simulated unless explicitly described otherwise. Phase 1 exposure totals are not Phase 2 totals.