DoVE — Evolutionary Design of Aerodynamic Vehicle Bodies
2014–2018 Evolutionary Algorithms 3D Shape Optimisation Computer Vision
DoVE — Development of Vehicle Exteriors
DoVE explored automated design of 3D aerodynamic objects. Using velomobile bodywork as a testbed, the project developed methods for evolving stable, lightweight, aerodynamically efficient shapes — without relying on expert engineering knowledge.
Why velomobiles?
The velomobile design community is driven by hobbyists and small companies whose trial-and-error process has produced highly non-intuitive but outstanding shapes (e.g., the Milan velomobile). This suggests a large, unexplored design space that conventional engineering intuition cannot reach — a perfect target for evolutionary exploration.
Technical approach
- Neural indirect encodings (CPPNs/HyperNEAT) — evolved to generate entire 3D geometries holistically, allowing non-local, highly complex shape changes
- Techniques originally developed for evolving large neural networks, repurposed for 3D form generation
- Integration with CFD evaluation for fitness assessment
DoVE-Tales: Real-World Drag Testing
A second strand extracted aerodynamic models from real-world footage:
- Tufts (yarn/wool indicators) attached to aerodynamic base shapes
- Computer vision analysis of tuft position, orientation, and deformation over time
- ML models deriving full quantitative aerodynamic characterisation from tuft behaviour
- Enabling realistic comparison tests and design analysis under actual road conditions
Partners
Prof. Alexander Asteroth (lead), Prof. Ernst Kruijff — H-BRS / TREE.