# PreferenceFlow: Test-Time Guidance of Flow-Matching Robot Policies from Human Interventions

arXiv: https://arxiv.org/abs/2609.36872 · HTML: https://mlsanity.com/paper/2609.36872/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Diffusion and flow policies for robotics, one of 14 directions in Diffusion models.
- Builds on 26 of its 29 references that are in the map.
- First posted on arXiv 29 Sep 2026.

## Placed in

- Diffusion models / [Diffusion and flow policies for robotics](https://mlsanity.com/diffusion/diffusion-and-flow-policies-for-robotics/)

## Builds on (most cited first, 26 of 26)

- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [Flow Matching for Generative Modeling](https://mlsanity.com/paper/2210.02747/) (2022)
- [Diffusion Policy: Visuomotor Policy Learning via Action Diffusion](https://mlsanity.com/paper/2303.04137/) (2023)
- [π0: A Vision-Language-Action Flow Model for General Robot Control](https://mlsanity.com/paper/2410.24164/) (2024)
- [π0.5: a Vision-Language-Action Model with Open-World Generalization](https://mlsanity.com/paper/2504.16054/) (2025)
- [GR00T N1: An Open Foundation Model for Generalist Humanoid Robots](https://mlsanity.com/paper/2503.14734/) (2025)
- [SmolVLA: A Vision-Language-Action Model for Affordable and Efficient Robotics](https://mlsanity.com/paper/2506.01844/) (2025)
- [World Action Models are Zero-shot Policies](https://mlsanity.com/paper/2602.15922/) (2026)
- [π*0.6: a VLA That Learns From Experience](https://mlsanity.com/paper/2511.14759/) (2025)
- [PEBBLE: Feedback-Efficient Interactive Reinforcement Learning via Relabeling Experience and Unsupervised Pre-training](https://mlsanity.com/paper/2106.05091/) (2021)
- [Cosmos Policy: Fine-Tuning Video Models for Visuomotor Control and Planning](https://mlsanity.com/paper/2601.16163/) (2026)
- [HG-DAgger: Interactive Imitation Learning with Human Experts](https://mlsanity.com/paper/1810.02890/) (2018)
- [Extrapolating Beyond Suboptimal Demonstrations via Inverse Reinforcement Learning from Observations](https://mlsanity.com/paper/1904.06387/) (2019)
- [Precise and Dexterous Robotic Manipulation via Human-in-the-Loop Reinforcement Learning](https://mlsanity.com/paper/2410.21845/) (2024)
- [Steering Your Diffusion Policy with Latent Space Reinforcement Learning](https://mlsanity.com/paper/2506.15799/) (2025)
- [Gemini Robotics 1.5: Pushing the Frontier of Generalist Robots with Advanced Embodied Reasoning, Thinking, and Motion Transfer](https://mlsanity.com/paper/2510.03342/) (2025)
- [RL Token: Bootstrapping Online RL with Vision-Language-Action Models](https://mlsanity.com/paper/2604.23073/) (2026)
- [OpenWAM: An Open, Modular Exploration Towards Systematic World-Action Model Pretraining](https://mlsanity.com/paper/2609.07398/) (2026)
- [Flex-π: A Multi-Stream World-Action Model with Compute Flexibility](https://mlsanity.com/paper/2608.10860/) (2026)
- [UniSteer: Unified Noise Steering for Efficient Human-Guided VLA Adaptation](https://mlsanity.com/paper/2605.10821/) (2026)
- [Test-Time Gradient Guidance of Flow Policies in Reinforcement Learning](https://mlsanity.com/paper/2606.11087/) (2026)
- [Improving Robotic Generalist Policies via Flow Reversal Steering](https://mlsanity.com/paper/2606.13675/) (2026)
- [FlowDAgger: Human-in-the-Loop Adaptation of Generative Robot Policies in Latent Space](https://mlsanity.com/paper/2607.08877/) (2026)
- [Modality-Autoregressive World-Action Models](https://mlsanity.com/paper/2609.17524/) (2026)
- [Steering Generative Robot Policies with Lexicographic Preferences](https://mlsanity.com/paper/2609.15014/) (2026)

## Abstract (arXiv)

Flow-matching policies can represent complex robot behaviors but remain susceptible to local errors under distribution shift at deployment. Many reinforcement learning approaches to policy improvement require reward signals that are difficult to specify or obtain in real-world manipulation. We present PreferenceFlow, a framework for improving a pretrained flow policy at test time without environment rewards or updates to the base policy. Human intervention chunks are paired with robot chunks generated from the same initial conditioning state to train a preference model. During infer- ence, we adopt the QGF sampling update, replacing its value gradient with the preference gradient evaluated at an estimated clean action. A gradient-cap loss penalizes excessive gradients on intervention pairs, while a zero-gradient loss discourages guidance near actions from expert demonstrations. On four real-world precision insertion tasks with a Franka robot, Pref- erenceFlow achieves a mean success rate of 90.5%, compared with 69% for the frozen policy. Ablations support the roles of gradient regularization and correctly ordered preference labels in the evaluated settings. These results demonstrate the utility of human interventions as local preference supervision for guiding frozen generative robot policies.