A ray-traced renderer for novel view synthesis that outruns the rasterisers.
Industrial PhD student · Zenseact × Chalmers
Bernardo Taveira
I work on generative and differentiable rendering for autonomous driving — novel view synthesis, scene generation, and sensor-realistic simulation.
A LEGO bulldozer from Mip-NeRF 360, ray traced live in your browser by . Drag to orbit.
News
-
VoroTracing is out as a preprint with code.
-
Scalable GPU Construction of 3D Voronoi and Power Diagrams has been accepted to SIGGRAPH 2026.
-
R3D2 has been accepted to the CVPR Workshop on Autonomous Driving.
Work
Building 3D Voronoi and power diagrams entirely on the GPU, at scale.
Diffusion that makes inserted 3D assets sit in a driving scene as if they belong.
Carrying state across LiDAR sweeps to detect objects the current frame barely sees.
About
I am a second-year industrial PhD student at Zenseact and Chalmers University of Technology, working with Fredrik Kahl and the Computer Vision group, funded by WASP. My research is on machine learning for autonomous driving — predominantly generative and differentiable rendering methods.
Before the PhD I was a hybrid system software engineer at Koenigsegg, and during my Master's I founded and led the autonomous systems group for Chalmers Formula Student's driverless racing project. I hold an MSc in Systems, Control and Mechatronics from Chalmers and a BSc in Electrical and Computer Engineering from Técnico, University of Lisbon.
Reach me at taveira@chalmers.se.
