To be able to interprete an image as a machine learning algorithm, the video stream created from this virtual world needed to be detailed enough. Otherwise the robot could interprete the data wrong and this would influence the learning of the robot.
To make a strawberry feel real, image scanning was used to produce large textures. After layering a decent amount of textures the strawberry came to life.
Backlighting and translucency
The leaves where made out of the same principles as the strawberry. One more difficulty was to create translucency through the leaves. A shader was written to create this effect with realtime lighting. You can check out the effect in the video below.
After the initial showcase of the 3D viewer, Octinion was convinced of the fact that these video stream images could be used to learn their algorithm much more efficiently.