Learning Multimodal Behaviors from Scratch with Diffusion Policy Gradient | Read Paper on Bytez