This looks like a cloud-enabled way to scale up agent-training for reinforcement learning, am I correct? If so, this could be a great enabler for developing robotic controls. I'm surprised that the article does not mention "reinforcement learning" at all.