Revisiting Parameter Redundancy in Vision-Language-Action Models: Insights from VLM-to-VLA Adaptation

2026-06-30

AI Summary

This work addresses the challenge that existing vision-language-action (VLA) models are highly parameterized and suffer severe performance degradation after pruning unless fine-tuned, masking critical issues of erroneously removed parameters. For the first time, it analyzes the spatial distribution of parameter changes from vision-language models (VLMs) to VLAs, introducing fine-tuning-free pruning as a diagnostic tool to uncover the causal relationship between parameter divergence and functional contribution. Building on these insights, the study proposes a multi-module joint pruning strategy that enables efficient model compression without requiring fine-tuning. Experiments on OpenVLA and π₀.₅ demonstrate 12%–30% parameter reduction while retaining approximately 90% of original performance—significantly outperforming existing methods, which exhibit substantial performance collapse under fine-tuning-free conditions.

Abstract

Vision-Language-Action (VLA) models have made significant strides in embodied intelligence by integrating the powerful representations of pre-trained Vision-Language Models (VLMs). However, the massive parameter scale of VLAs imposes a heavy computational burden, and these models exhibit extreme sensitivity to parameter pruning. Current paradigms often treat the resulting performance degradation as inevitable, relying on fine-tuning or low-rank corrections to recover efficacy. We challenge this convention by questioning whether the removed parameters are truly redundant if VLA pruning necessitates performance recovery to be effective, or if this paradigm masks the indiscriminate pruning of critical parameters. We revisit parameter redundancy through the lens of VLM-to-VLA adaptation, first quantifying the spatial distribution of parameter divergence during adaptation to reveal structured patterns across different modules. Subsequently, we introduce controlled pruning as a diagnostic probe: by comparing the direct impact of removing different parameter subsets on VLA performance without any fine-tuning, we establish a causal link between adaptation-induced divergence signals and functional contributions. Based on the discovered modular heterogeneities, we design a multi-module joint pruning scheme. Evaluations on the LIBERO benchmark demonstrate that our approach reduces the parameters of OpenVLA and $π_{0.5}$ by 12\%--30\% while maintaining approximately 90\% of the original performance without any post-pruning recovery. In contrast, existing parameter pruning criteria result in total performance collapse when evaluated under the same recovery-free constraints. Our study reveals the parameter evolution mechanism in VLA adaptation and provides a new path for deploying efficient, robust robotic policies in resource-constrained environments.

Problem

• Vision-Language-Action models
• parameter redundancy
• model pruning
• VLM-to-VLA adaptation
• embodied intelligence

Innovation

• parameter pruning
• vision-language-action models
• module heterogeneity
• controlled pruning
• adaptation-induced divergence