RFPO: Rectified Flow Policy Optimization for Embodied Control
Flow-based policies provide an expressive framework for continuous robot control, but their iterative ODE integration incurs substantial inference cost.
Key points
- To address this problem, we introduce RFPO, a flow-policy optimization framework for reliable few-step execution.
- Reward-aware online Reflow rectifies student-induced transport paths during on-policy learning, making the resulting policy more robust to coarse integration.
- A frozen Gaussian PPO controller supplies complementary action-space supervision at full and intermediate integration budgets, while the deployed policy remains a single flow student executed with one Euler step.
- Real-robot experiments further validate stable one-step locomotion.
Sources (1)
- [1]RFPO: Rectified Flow Policy Optimization for Embodied ControlarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 7, 05:24 PM
Flow-based policies provide an expressive framework for continuous robot control, but their iterative ODE integration incurs substantial inference cost.
To address this problem, we introduce RFPO, a flow-policy optimization framework for reliable few-step execution.
Extractive summary: sentences quoted from the sources.