This project aimed to predict ambient occlusion values based on the joint angles. The idea was put the mesh in arbitrary poses and compute the raytraced ambient occlusion with OptiX for randomly sampled uvs along the mesh. Then, by passing the relative joint angles and the uv into an MLP, the occlusion values would change with respect to the thetas. In my case, I modified an arm obj file and sampled 50 poses ranging from a theta value of 0-100 degrees with 10 samples per mesh triangle. The results were disappointing but I am hopeful that with more data and tuning this will prove interesting.