I am working on robots that see the world through two cameras. For development, we’d love to use the existing vision pipeline (written in CUDA and C++), but replace the cameras with 3d scene renderers (So we can essentially run unit tests).