UC Berkeley's QF3 Enables Stable RL Fine-Tuning of Flow Policies

UC Berkeley and Amazon FAR released QF3 (Fast Flow RL with Filtered Q-Gradients), which stabilizes RL fine-tuning of flow-based robot policies with a single change to the actor update. The project page and paper (arXiv 2610.08789) include interactive explanations and all experiment videos.

2026-10-13 ~ 2026-10-13 · 2 related posts

• Berkeley's QF3 makes RL work for flow policies on real robots with one clipped gradient tweak — andrew_n_carr · 2026-10-13
• QF3 project page live: interactive explainer for fast flow RL on robots — andrew_n_carr · 2026-10-13