Project 02 | AI driven character behaviour system
AI Driven Behaviour System for Realistic Game Characters
- Field
- AI and game technology
- Scope
- TÜBİTAK 1507 | Project No 7250714
- Developed by
- Rysesoft
- agents at 60 FPS
- 500
- Lazy Theta* over 1,000 queries
- 1.9 ms
- swarm cost, down from O(n²)
- O(n)
- technology readiness level
- 4 → 8
Goal
We wanted game characters that find their way in three dimensions, move as a swarm and decide for themselves, in a system that works in any game.
The Smart NPC System brings real time pathfinding, swarm movement and decision making for game characters together in three modules. It does not depend on one game. The enemy drone team of Starship Crafter runs on it.
We developed it on Unity between July 2025 and June 2026 with the support of TÜBİTAK 1507, in the project "Gerçekçi Oyun Karakterleri Davranışları için YZ Destekli Gerçek Zamanlı Yol Bulma, Sürü Hareketleri, Durum ve Karar Mekanizmalarının Tasarlanması ve Geliştirilmesi" (No 7250714).
Project adviser
- Prof. Dr. Tolga Kurtuluş Çapın
Monitoring reviewer
- Prof. Dr. Cemal Köse
How the System Works
- 01
Pathfinding in 3D
Voxel based scene mapping with Octree and SVO structures lets characters find their way in three dimensional space.
- 02
Four Algorithms Compared
A*, Dijkstra, Theta* and Lazy Theta* were implemented and measured side by side on a 50×50×10 grid.
- 03
Swarm Movement
Spatial hashing turns the swarm calculation from O(n²) into O(n), so 500 agents run at 60 FPS on an RTX 5060.
- 04
Hybrid Decisions
A behaviour tree and a decision tree work together, and profiles (ScriptableObject) shape each character.
- 05
Targets Passed
A swarm of 32 characters, 64 agents active at the same time and more than 8 states, all above the project's targets.
- 06
In Starship Crafter
The enemy drone team of Starship Crafter runs on the system, which is also being prepared as a Unity package of its own.
