Based on the decison on #5, either a custom component that can render raw keypoint sequences needs to written, or a FBX file containing all keypoints as animation nodes needs to created by the ML environment.
We can use this separate repository: https://github.com/deep-dance/visualizer. I'll push a prototype written in Rust. Maybe it'll help to get a better understanding of what we need.
Based on the decison on #5, either a custom component that can render raw keypoint sequences needs to written, or a FBX file containing all keypoints as animation nodes needs to created by the ML environment.