Umeå universitets logga

Recent advancements in digital physics simulations, particularly in game engines like Unity and Unreal Engine, have enabled industries such as construction, mining, and agriculture to develop training platforms and prototyping tools. Some of these simulations incorporate terrain dynamics where soil is often modeled as collections of particles which may interact. However, challenges arise when attempting to enhance visual fidelity by scaling the number of particles to the millions, due to computational limitations. A primary method for increasing the particle count is called particle upsampling, a technique that increases the number of visual particles while maintaining good performance. Traditional game engines prioritize performance and visuals over physical accuracy, often requiring plugins or external solutions for more physically accurate simulations.

This study evaluates and compares the integration of a hashmap-based particle upsampling solution in Unity and Unreal Engine where the coarse particles are generated by the physics engine AGX Dynamics. It examines three core aspects: performance, code complexity, and visual fidelity. Performance metrics, in terms of GPU execution time, show that Unreal Engine outperforms Unity, provided that initial buffers are properly sized to avoid dynamic buffer resizing. Code complexity analysis shows both implementations are maintainable, with Unreal's reliance on Niagara simplifying buffer management, while Unity offers more transparent control. Visual fidelity evaluations highlight improvements in particle dynamics such as movement and particle count, though issues such as particle extinction and shadow inaccuracies persist, particularly in Unreal Engine.

The findings demonstrate that both Unity and Unreal Engine are capable of running particle upsampling solutions effectively, each with specific trade-offs in terms of implementation complexity, performance characteristics, and visual outcomes. The fact that both engines deliver strong results across all three evaluation metrics underscores the practicality and robustness of the upsampling method itself. Ultimately, the choice of engine and the viability of integration depend on the priorities of the developers. However, this research provides a clear indication that these types of solutions are well-suited for real-time contexts such as game engines.