There is no real Training-Code yet. We need an environment which allows fast multiagent learning. A C++/Bullet-Implementation might be a good solution. We are also currently evaluating Unity as an Enviroment, but tbh i prefer to have a clean and tight C++ implementation instead of Mono/C#. On the other hand Unity would allow us to have far more realistic (virtual) VideoInputs for CV-Training. Not sure about this, would be great to have feedback here from you guys. I also started to create a simple OpenAI-Gym-Environment, but have not finished it yet :/