Lowering the Reality Gap in Aggregate Programs Validation: Running Collektive Over Unity

Modern computing increasingly relies on Complex Adaptive Systems (CAS) (i.e. large ensembles of autonomous agents whose collective behavior emerges from local interactions). Aggregate Computing (AC) provides a rigorous framework for programming such systems, yet validating AC algorithms before physical deployment remains challenging. Classical simulators prioritize scalability but sacrifice physical realism, while high-fidelity robotic simulators are computationally expensive and poorly suited to AC workflows. This mismatch, the reality gap, risks producing algorithms that succeed in simulation but fail in the field. This thesis presents Collektivity, a 3D simulator for CAS that bridges the Unity game engine with Collektive, a Kotlin-based AC library. Unity supplies realistic physics, collision detection and spatial partitioning, while Collektive handles aggregate logic. The two runtimes communicate through a Foreign Function Interface (FFI) layer with Protocol Buffers as the shared data model. To support Collektivity as a distributable library, a Unity Package Template was also developed, integrating continuous integration, automated semantic versioning, static analysis and artifact signing. A benchmark comparing FFI against socket-based communication showed that FFI is on average 450 times faster, with worst-case speedups exceeding 700 times, enabling the 20Hz execution frequency required by AC. A case study in environment-aware gradient ascent validated the integration across a minimal and a high-fidelity 3D environment, sustaining real-time performance with 100 nodes on consumer hardware. These results demonstrate that game engines are viable high-fidelity platforms for CAS validation, concretely narrowing the reality gap between algorithmic design and physical deployment.