Unified Noise Steering for Efficient Human-Guided VLA Adaptation

Summary

Diffusion-based vision-language-action (VLA) models have emerged as strong priors for robotic manipulation, yet adapting them to real-world distributions remains challenging.

Abstract Summary

Adapting Vision-Language-Action (VLA) models to specific user preferences or task requirements typically requires expensive fine-tuning. This paper introduces Unified Noise Steering, a lightweight method that guides VLA model behavior by steering the diffusion noise process during inference rather t... The method demonstrates significant improvements over existing approaches, providing both theoretical insights and practical benefits for real-world deployment. Comprehensive experiments validate the effectiveness of the proposed approach across diverse scenarios and task settings.

Key Points

Proposes Unified Noise Steering for Efficient Human-Guided VLA Adaptation, a novel approach for vla in robotics.
Addresses key limitations in existing methods through innovative architecture design.
Demonstrates strong empirical results on standard benchmarks and real-world evaluations.
Introduces novel training methodology that improves generalization and sample efficiency.
Provides comprehensive analysis of failure modes and ablation studies.

Problem Formulation

We consider VLA policies parameterized as conditional flow matching models.

Given a state $s$, the policy maps an initial noise variable $z_{0}\sim\mathcal{N}(0,I)\in\mathbb{R}^{d}$ to an action chunk $a\in\mathbb{R}^{d}$ by integrating a state-conditioned velocity field $v_{\theta}(z_{t},t,s)$, where $\theta$ is the parameter of the network: At each flow step $k$, the network predicts a velocity update conditioned on the current noise, the flow time, and the state.

Problem Formulation — Unified Noise Steering for Efficient Human-Guided VLA Adaptation — Figure 2 : Overview of UniSteer framework. The noise actor maps current states into noise variables and generates actions through the frozen decoder. Autonomous rollouts are stored in the RL buffer, whereas human takeover actions are inverted into noise space and collected in both the demo buffer and the RL buffer. The two buffers jointly provide training signals for the critic and the actor.

Method

Given the current state $s$ and a human-provided corrective action chunk $a^{h}$, our goal is to recover a noise target $\hat{z}$ such that, the frozen flow-matching policy can decode it into an action chunk close to $a^{h}$. Ideally, the desired noise target is the inverse image $G_{\theta}(s,\cdot)^{-1}(a^{h})$.

However, since this noise-to-action mapping is realized through a multi-step nonlinear flow decoder, directly inverting it is difficult.

Method — Unified Noise Steering for Efficient Human-Guided VLA Adaptation — Figure 3 : Overview of all real-world experimental tasks.

Experiment

We evaluate our method on four real-world manipulation tasks: pick up spoon , stack blocks , insert square , and fold towel , as shown in Figure 3 . These tasks cover a broad range of manipulation skills, including pick-and-place, precise contact-rich operations, and deformable-object manipulation.

Together, they provide a compact benchmark for evaluating both broad spatial generalization and fine-grained manipulation refinement under distribution shift.

Experiment — Unified Noise Steering for Efficient Human-Guided VLA Adaptation — (a) Trajectory composition per adaptation round.

Sources and Demonstrations

Figures are reproduced from Lu et al. See the full paper for experimental details and the project page, when the paper links one, for demonstrations and videos.