Swarm AI Toolkit – Navigation Bundle

Engine & Compatibility

• Unity 2021 LTS or newer recommended
• Compatible with Built-in Render Pipeline, URP, and HDRP
• No render-pipeline-specific runtime dependencies
• No required third-party runtime libraries
• PC, console, and mobile supported, depending on agent count and settings

Architecture Overview
The Navigation Bundle consists of three modular movement systems:

1. Swarm Core - Agent Simulation Layer

• Centralized system-driven update loop
• Lightweight agent movement
• 2D planar and 3D spatial query support
• Spatial hash grid for neighbor and proximity queries
• Physics-free movement model
• Radius-based agent representation
• Managed fallback path
• Burst/job-system groundwork for movement, spatial hash rebuilds, and separation
• Designed for large agent counts

2. Flow Fields - Global Navigation Layer

• Grid-based cost field generation
• Shared direction fields for groups of agents
• Single-layer and layered flow-field navigation
• Multi-target direction field support
• Dynamic target updates
• Static and dynamic cost stamping
• Dynamic cost contributors
• Terrain, raycast, fixed-altitude, and baked-heightmap height workflows
• No per-agent pathfinding
• Designed for mass movement scenarios

3. Steering + Avoidance - Local Movement Layer

• Predictive agent-agent collision avoidance
• ORCA 2D reciprocal avoidance option
• ORCA 3D-style avoidance for flying agents
• Continuum-style crowd avoidance option
• Dynamic obstacle avoidance
• Cost-field and obstacle-field avoidance
• Steering behaviours including seek, flee, arrive, follow, pursue, wander, orbit, density, formation, boids, separation, cohesion, and alignment
• Velocity blending, smoothing, weights, and priorities
• Avoidance urgency reporting for stable behavior under collision pressure

Movement Pipeline

• Global direction from Flow Fields
• Local avoidance adjustment
• Steering blend and behavior weighting
• Swarm Core movement integration
• Optional terrain or fixed-altitude projection
• Final agent position and velocity

Performance Characteristics

• Designed for large-scale simulations with hundreds to thousands of agents
• Performance depends on:Agent count
  Flow-field grid resolution
  Neighbor query radius
  Max neighbor count
  Avoidance module settings
  Update rate and rebuild intervals
• Non-alloc query paths used where possible
• Spatial hashing reduces broad-phase query cost
• Dynamic flow-field rebuilds can be skipped when contributors have not changed
• No Rigidbody components required for default movement
• No Unity NavMesh dependency required
• System-driven update model with lightweight agent components

Movement Model

• Supports planar XZ movement
• Supports terrain-following ground agents
• Supports fixed-altitude and flying-style agents
• Supports 3D neighbor queries for flying swarms
• Configurable radius, speed, mass, steering weights, and priority
• Height projection can be handled through terrain, raycast, fixed altitude, or heightmap sampling

Integration

• Swarm Core can run standalone
• Flow Fields integrates directly with Swarm Core
• Steering + Avoidance can run standalone or through Swarm Core adapters
• Flow Field cost data can be used by avoidance modules
• Designed for modular extension
• Compatible with custom AI decision systems
• Can be combined with separate Perception and Decision modules

Limitations

• Flow Fields are grid-based and primarily XZ-oriented for global navigation
• Flying support is suitable for layered/fixed-altitude and steering-driven vertical movement, not full volumetric 3D pathfinding
• No built-in animation controller
• No built-in combat logic
• No required NavMesh crowd simulation
• Very large scenes require tuning grid size, update rates, neighbor radius, and avoidance settings.