Reinforcement Learning via Fixed-point Flow Matching
To address this limitation, we propose fixed-point flow matching (FFM), a data-free procedure for training and fine-tuning flow policies in RL.
ProofPaper ↗
Key points
- In maximum entropy and behavior-constrained reinforcement learning (RL), the optimal policy is given by a Boltzmann distribution.
- We show that the optimal policy is a fixed point of this iteration.
- Through extensive experiments, we demonstrate that FFM effectively fine-tunes pretrained flow policies in the offline-to-online setting and enables efficient and accurate training of flow policies from scratch in online RL.
- These results establish fixed-point flow matching as a principled approach for learning expressive flow-based policies without requiring training data.
Sources (1)
- [1]Reinforcement Learning via Fixed-point Flow MatchingarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 9, 01:15 AM
To address this limitation, we propose fixed-point flow matching (FFM), a data-free procedure for training and fine-tuning flow policies in RL.
In maximum entropy and behavior-constrained reinforcement learning (RL), the optimal policy is given by a Boltzmann distribution.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 9, 2026Cross-Behavior Policy Optimization: Unified Learning from Offline Demonstrations and On- and Off-Policy Rollouts via a Shared Regression Target
- Oct 9, 2026Learning Representations to Guide Pixel Diffusion Transformers
- Oct 9, 2026Rethinking Relative Policy Differences Transfer of On-Policy Distillation
- Oct 8, 2026StereoEngine: Scaling Stereo Bimanual Robot Data without Stereo Cameras
- Oct 8, 2026Visual-Wi-Fi Navigation to Signal-Emitting Devices
- Oct 8, 2026Zatom-2: Multitask Pretraining on Atomistic Data for Generative Modeling across Domains