InSpire: Vision-Language-Action Models with Intrinsic Spatial Reasoning

TL;DR

InSpire enhances vision-language-action models for robotics by improving spatial reasoning, reducing irrelevant visual correlations, and boosting generalization without additional training or data, demonstrated through extensive experiments.

Contribution

The paper introduces InSpire, a simple plugin that improves spatial reasoning in VLAs, addressing spurious correlations and enhancing generalization in robotic tasks.

Findings

InSpire significantly improves task success rates in simulation and real-world tests.

It effectively reduces reliance on irrelevant visual features.

The approach requires no extra training data or large model interactions.

Abstract

Leveraging pretrained Vision-Language Models (VLMs) to map language instruction and visual observations to raw low-level actions, Vision-Language-Action models (VLAs) hold great promise for achieving general-purpose robotic systems. Despite their advancements, existing VLAs tend to spuriously correlate task-irrelevant visual features with actions, limiting their generalization capacity beyond the training data. To tackle this challenge, we propose Intrinsic Spatial Reasoning (InSpire), a simple yet effective approach that mitigates the adverse effects of spurious correlations by boosting the spatial reasoning ability of VLAs. Specifically, InSpire redirects the VLA's attention to task-relevant factors by prepending the question "In which direction is the [object] relative to the robot?" to the language instruction and aligning the answer "right/left/up/down/front/back/grasped" and predicted actions with ground-truth. Notably, InSpire can be used as a plugin to enhance existing autoregressive VLAs, requiring no extra training data or interaction with other large models. Extensive experimental results in both simulation and real-world environments demonstrate the effectiveness and flexibility of our approach.