FLAG: Flow Policy MaxEnt-RL by Latent Augmented Guidance
FLAG: Flow Policy MaxEnt-RL by Latent Augmented Guidance
This research presents FLAG (Flow policy with Latent-Augmented Guidance), a method that augments the state space with a flow latent variable to optimize a consistent proxy for maximum entropy reinforcement learning. By localizing the sampling region, FLAG mitigates importance weight collapse common in generative policies, enabling expressive policy optimization with limited samples and scaling to high-dimensional control tasks. Empirical results show it achieves state-of-the-art performance among expressive-policy MaxEnt online RL methods on state-based high-dimensional control benchmarks.
The paper introduces FLAG, a method combining flow-based latent augmentation with maximum entropy reinforcement learning to address importance weight collapse in high-dimensional action spaces, achieving state-of-the-art performance on relevant benchmarks.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
Source：arXiv Robotics — research abstracts · arxiv.org