As part of my PhD research at RMIT Architecture, I’ve been developing a set of design tools that explore the relationship between gestural input and multi-agent simulation, using emergence to generate intricate geometry. Below is an example mesh running a multi-agent simulation at ~1 million vertices in realtime on an RTX 2080 Ti:
In this example I’m experimenting with mouse momentum to steer the forces. I’ve released a public version of the plugin if you’d like to try it:
https://www.food4rhino.com/en/app/aurora
Windows only at the moment (Sorry!) - I need to figure out how to get this running on metal. Or to just swap to a cpu friendly fallback.
Plugin Overview Video:
https://vimeo.com/manage/videos/1171461518
Getting multi-agent algorithms to run in realtime is challenging - especially with n-body style interactions, where you need spatial search and often heavy mesh reconstruction per frame. Aurora tackles this in a few key ways:
-
The simulation runs asynchronously, so it doesn’t lock the main UI thread.
-
GPU-native geometry classes handle topology changes.
-
Most simulation/compute runs GPU-side.
-
Updated vertex/normal/colour data is moved through the pipeline efficiently.
-
Currently the display path is GPU → CPU → GPU; I’m working toward pointing Rhino’s display buffers at a GPU “staging” area to avoid the round trip.
In the current build I’ve also been testing two approaches for computing n-body motion vectors:
-
Spatial Hash: points are hashed into a grid of cell size x, which accelerates neighbour queries.
-
Sparse Eulerian Grid: borrows from fluid simulation ideas by computing vectors as a pressure field across sparse grid “bricks.”
Overall, the biggest speed gains have come from tightening the display pipeline and improving the spatial search logic.
I am open to any feedback/suggestions to improve the plugin.
We’ve used this workflow to deliver several recent research projects at the RMIT Tectonic Formation Lab including:
3D Printed Sandstone & Polymer Pavilion:
The project was designed by the Tectonic Formation Lab in collaboration with SEU Architecture | Lab of Architectural Algorithms & Applications (Lab.AAA). The installation comprises of two interdependent printed systems: a primary sand-printed structural form produced through large-scale binder-jet fabrication; and a secondary layer of translucent, robotically extruded polymer louvered fins.
Full project credits
3D Printed Polymer Ceiling Feature (Advanced Manufacturing Week Expo)
This large-scale 3D-printed polymer ceiling installation translates generative ornament into a fabricated architectural surface. Aurora was used to generate intricate, flowing detail through a realtime, mesh-based behavioural workflow - producing a dense field of striation and relief that reads as both geological and architectural.
Rather than treating fabrication as a final step, the form was developed in direct response to manufacturing constraints. The surface topology was iteratively sculpted and refined around draft-angle limits to prevent print failure and support repeatable production - keeping complex ornament materially grounded and buildable at scale.
