DynamicPlanningAI (DP_AI)

Engine-agnostic, deterministic Goal-Oriented Action Planning (GOAP) framework for tactical AI agents. Written in C# targeting .NET 10, with bounded capacities, Power-of-Ten style constraints, and host-facing contracts that do not depend on any game engine.

Naming

Form	Meaning
DynamicPlanningAI	Full product / solution / assembly / namespace name
DP_AI	Accepted short form of DynamicPlanningAI
DP	Always means DynamicPlanning (never a standalone product name)

Task identifiers use the DP- prefix (for example DP-PLN-001) because DP means DynamicPlanning. See docs/NAMING.md.

Fidelity disclaimer

DynamicPlanningAI is inspired by publicly documented F.E.A.R.-era GOAP / tactical AI principles. It is original engineering, not a reverse-engineered or proprietary reproduction of any commercial AI codebase. See NOTICE.md and docs/PUBLIC_SOURCE_FIDELITY.md.

Features

• Backward regression A* planner over a bit-mask symbolic world state
• Incremental planning with per-step expansion budgets and explicit PlannerStatus outcomes
• Goal arbitration selecting among competing agent goals each tick
• Action execution lifecycle with volatile precondition checks and typed failure reasons
• Perception and working memory feeding quantized world facts
• Target / weapon selection, tactical points, and cover reservation
• Navigation host contracts (engine-agnostic path queries)
• Squad coordination and semantic communication intents
• Deterministic ticks via host-provided AiTick (no wall-clock reads)
• Frozen runtime path with allocation discipline and audit tooling
• Console sample scenarios (e.g. BasicAttack) for regression demos

Solution layout

Project	Role
DynamicPlanningAI.Abstractions	Contracts, identifiers, limits, domain enums
DynamicPlanningAI.Runtime	Planner, memory, cover, squad, execution
DynamicPlanningAI.Configuration	Validated immutable config loading
DynamicPlanningAI.Diagnostics	Tracing, formatters, contract helpers
DynamicPlanningAI.Sample	Deterministic console simulation
DynamicPlanningAI.Audit	Power-of-Ten / allocation-path source audit
tests/*	Unit, integration, simulation, architecture tests

Dependency direction:
Abstractions ← Runtime ← Configuration ← Sample
Abstractions ← Diagnostics (also referenced by Sample)

Build, test, and sample commands

export PATH="$HOME/.dotnet:$PATH"

dotnet restore
dotnet build --configuration Release
dotnet test --configuration Release --no-build
dotnet format --verify-no-changes
dotnet run --project tools/DynamicPlanningAI.Audit --configuration Release
dotnet run --project src/DynamicPlanningAI.Sample --configuration Release -- --scenario BasicAttack

Requires the .NET SDK version pinned in global.json (10.0.x).

Architecture overview

flowchart TB
  Host[Host / Engine Adapter]
  Sample[DynamicPlanningAI.Sample]
  Config[DynamicPlanningAI.Configuration]
  Runtime[DynamicPlanningAI.Runtime]
  Diag[DynamicPlanningAI.Diagnostics]
  Abs[DynamicPlanningAI.Abstractions]

  Host --> Runtime
  Sample --> Config
  Sample --> Runtime
  Sample --> Diag
  Config --> Runtime
  Runtime --> Abs
  Diag --> Abs
  Config --> Abs

  subgraph RuntimeTick[Per-tick pipeline]
    Perc[Perception]
    Mem[Working Memory]
    Goal[Goal Arbitration]
    Plan[GOAP Planner]
    Exec[Action Execution]
    Squad[Squad Coordinator]
    Comm[Communication]
    Perc --> Mem --> Goal --> Plan --> Exec
    Squad --> Goal
    Comm --> Mem
  end

  Runtime --- RuntimeTick

Loading

See docs/ARCHITECTURE.md for subsystem boundaries and lifecycle states.

Documentation index

Document	Topic
docs/ARCHITECTURE.md	Layers, lifecycle, tick pipeline
docs/PUBLIC_SOURCE_FIDELITY.md	Inspiration vs. originality
docs/POWER_OF_TEN_COMPLIANCE.md	NASA Power of Ten → C#
docs/GOAP_PLANNER.md	Backward regression A*
docs/WORLD_STATE_MODEL.md	Symbolic facts and ownership
docs/GOAL_ARBITRATION.md	Goal selection
docs/ACTION_EXECUTION.md	Action lifecycle
docs/PERCEPTION_AND_MEMORY.md	Sensors and working memory
docs/TARGET_AND_WEAPON_SELECTION.md	Focus and loadout
docs/TACTICAL_POINTS_AND_COVER.md	Cover / flanking demo
docs/NAVIGATION_INTEGRATION.md	Path host contracts
docs/SQUAD_COORDINATION.md	Squad behaviors and orders
docs/COMMUNICATION_SYSTEM.md	Semantic intents
docs/DETERMINISM.md	Tick and RNG rules
docs/MEMORY_AND_ALLOCATIONS.md	Bounds and freeze path
docs/CONFIGURATION.md	Config validation
docs/ENGINE_INTEGRATION_GUIDE.md	Hosting in any engine
docs/EXTENDING_GOALS_AND_ACTIONS.md	Authoring extensions
docs/DEBUGGING_AND_TRACING.md	Diagnostics
docs/TESTING.md	Test strategy
docs/PERFORMANCE.md	Budgets and profiling
docs/FAILURE_MODES.md	Explicit failure handling
docs/DESIGN_DECISIONS.md	ADRs
docs/REQUIREMENTS_TRACEABILITY.md	REQ → files
docs/BACKLOG.md	Task backlog

Hard limits (summary)

Absolute ceilings live in AiHardLimits (Abstractions). Examples:

Limit	Value
Agents	64
Squads	16
Goals per agent	32
Plan length	16
Planner nodes	512
World facts	64
Tactical points	256

Runtime configuration may only select values at or below these ceilings.

Limitations

• Runtime, configuration loaders, sample scenarios, and audit rules are under active implementation; Abstractions contracts and limits are the stable foundation today. Track status in docs/REQUIREMENTS_TRACEABILITY.md.
• Planning is symbolic only; continuous combat math belongs in the host or in pre-planning quantizers.
• Navigation mesh queries are host-provided; the library does not embed a pathfinder.
• Not a full animation, physics, or networking stack.
• Single-threaded tick assumption: the host must serialize AiRuntime ticks.

License

MIT — see LICENSE.md. Third-party and inspiration notices: NOTICE.md.