# VISTA: Enhancing Visual Conditioning via Track-Following Preference Optimization in Vision-Language-Action Models

Source: [VISTA: Enhancing Visual Conditioning via Track-Following Preference Optimization in Vision-Language-Action Models](<https://www.sota2.com/research/paper/69857565c97a0b3e3b5d5ae6>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-02-07T05:00:04.109Z

Authors: Yiye Chen, Yanan Jian, Xiaoyi Dong, Shuxin Cao, Jing Wu, Patricio Vela, Benjamin E. Lundell, Dongdong Chen

Published: 2026-02-04T20:59:29.000Z

## Abstract

Vision-Language-Action (VLA) models have demonstrated strong performance across a wide range of robotic manipulation tasks. Despite the success, extending large pretrained Vision-Language Models (VLMs) to the action space can induce vision-action misalignment, where action predictions exhibit weak dependence on the current visual state, leading to unreliable action outputs. In this work, we study VLA models through the lens of visual conditioning and empirically show that successful rollouts consistently exhibit stronger visual dependence than failed ones. Motivated by this observation, we propose a training framework that explicitly strengthens visual conditioning in VLA models. Our approach first aligns action prediction with visual input via preference optimization on a track-following surrogate task, and then transfers the enhanced alignment to instruction-following task through latent-space distillation during supervised finetuning. Without introducing architectural modifications or additional data collection, our method improves both visual conditioning and task performance for discrete OpenVLA, and further yields consistent gains when extended to the continuous OpenVLA-OFT setting. Project website: https://vista-vla.github.io/ .

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Long-Horizon Multi-Task Language Control on CALVIN ABC→D (test)](<https://www.sota2.com/research/sota/long-horizon-multi-task-language-control-on-calvin-abc-d-test.md>): Long-Horizon Multi-Task Language Control; CALVIN ABC→D (test); Seq Success (1): 94.8.

- [Robot Manipulation on LIBERO](<https://www.sota2.com/research/sota/robot-manipulation-on-libero.md>): Robot Manipulation; LIBERO; Object Achievement: 91.6.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2602.05049>)

[PDF](<https://arxiv.org/pdf/2602.05049.pdf>)