Policy Optimization for Generative Models
We study policy optimization for generative models, beginning with D2PPO and extending from diffusion policies to one-step flow policies, multi-agent learning, and inference-time guidance and planning.
| Project | Research direction |
|---|---|
| D2PPO | Diffusion policy optimization with dispersive loss |
| DM1 | MeanFlow with dispersive regularization for one-step manipulation |
| CoFlow | Coordinated few-step flow for offline multi-agent decision-making |
| OGPO | One-step generative policy optimization for real-time robot control |
| MA-FPPO | Multi-agent flow-pretrained policy optimization |
| G2MAF | Test-time gradient guidance for multi-agent flow policies |
| MA-WAM | Multi-agent world-action modeling for test-time planning |
Explore the projects and available implementations in POInfra.
Maintained by Guowei Zou, Sun Yat-sen University.