It’s flattering to see people trying to work out how we did this!
All of the schematic design was done without any surface modelling at all, using really simple techniques that as Michael mentions we’ve all seen a million times before. We were really just interpolating curves through vector fields (using Nursery Nursery | Food4Rhino though I haven’t maintained this for ages). The nice thing about this design approach is you can generate really complex topologies really quickly. No underlying mesh, or even agenty boidy stuff needed.
After we were thrown under the bus and actually had to build the thing we built a surface model of the design and then used this as as a quick way of generating strips that didn’t intersect one another.
The complex (and interesting) parts of the design process all come from trying to simulate how the timber board will twist as you try to bend it into a particular curve, as this gives you both the design language of the pavilion (when strips overlap you read surface, when they don’t you read lines) and also the geometry needed to generate steel brackets that connect the strips together. We ended up using kangaroo to simulate both of these things. It was fun trying to build in all of the constraints for the steel brackets in particular - they needed to unroll to a straight bar so we could use standard flat bar to make them, needed to have minimal connections to surfaces, needed to have max bend angles so we could bend them by hand, couldn’t intersect anything (obviously) and needed to form geometry that would sit behind the strips where possible.
Then everything was built with the HoloLens and Fologram. A video showing a few of the AR tricks behind the project is here: Steampunk Pavilion