Find your way through DEC-MAPF¶

Documentation index · Architecture · Contributing

Start with a small, inspectable result. Choose the interface that fits your work; both use the same application services and saved evidence.

flowchart LR
    A[Scenario + explicit settings] --> B[Preview and frozen manifest]
    B --> C[Supervised solver processes]
    C --> D[Independent trajectory validation]
    D --> E[Journal and run artifacts]
    E --> F[CLI outcome tables and analysis]
    E --> G[Optional GUI replay and export]

Choose a starting point¶

You want to…	Begin with	Check before drawing a conclusion
Understand one simulation	GUI walkthrough	Actual executed moves and the selected method's recorded diagnostics
Run a matrix unattended	Headless tutorial	Every planned case, explicit budgets and attempt selection
Inspect many saved results	Run library	Filters, page count and matching configuration
Compare decentralized and centralized methods	Method comparison, metrics and experiments	Common valid solved instances for costs; full denominator for success
Add a solver, strategy or coordination protocol	Extension contracts	Independent validation, declared information/control model and method identity
Diagnose an installation	`mapf doctor` and installation	Required extras, built assets and workspace schema

Repository contents and ownership¶

Path	Responsibility	When to use it
`src/mapf/core/`	Grid, paths, settings, world and independent validator	Reason about discrete problem semantics
`src/mapf/agents/`, `src/mapf/negotiation/`	Agent strategies and token protocol	Trace a negotiation decision
`src/mapf/solvers/`	Solver adapters and independent validation boundary	Run or qualify an algorithm
`src/mapf/application/`	Presets, plans, supervision, journals, diagnostics and exports	Build a GUI-independent workflow
`src/mapf/gui/`, `frontend/`	HTTP adapter and optional visual workspace	Inspect the same evidence interactively
`examples/`	Small declared specifications and runnable Python examples	Learn or build a minimal reproduction
`tests/`, `frontend/tests/`	Behavioral and browser acceptance witnesses	Regress a concrete failure
`docs/`	Tutorials, concepts, contracts, gallery and provenance	Interpret features and results
`benchmarks/`	Software regression and profiling fixtures	Measure declared engineering workloads; use `mapf batch` for your studies
`runs/`, `data/workspace/`	Ignored local outputs, never packaged research evidence	Keep your own manifests and journals

What a picture can tell you¶

The gallery shows validated modern solutions from two article settings: an empty 16×16 grid with 80 agents and a 32×32 grid with approximately 20% obstacles and 80 agents. It includes selected-agent local heat, recorded observations, commitments and an executed route. Each caption explains the solver, size and evidence. The gallery receipt records run identities and image hashes.

Use a clean position/goal view for a large-map overview, then select one agent before adding paths, field of view and reservations. Use synchronized replay for a matched comparison. A visually attractive, collision-free example does not establish scaling, runtime superiority or reproduction of published scores.

Learning from other tools¶

Related tools explains the engineering patterns that informed this repository. Those links are references, not new solver dependencies or claims of comparable experimental tasks.