Join me on my journey towards physics-based 3D animation generation using neural networks!
This year I've been working on setting up an animating humanoid in the physics sim program Pybullet.
Today I got the animation data pipeline in place!
The cool part is that a physics-capable robot in Pybullet is doing it.
Robotics programs don't really support animation out of the box - you gotta write your own way to do everything, from stepping time forward, to converting animation data and applying it to the robot's joints. (For me ~2500 lines of python)
Now that I have this framework, I can finally start running machine learning on top of it!