Beyond Attention Magnitude: Leveraging Inter-layer Rank Consistency for Efficient Vision-Language-Action Models

TL;DR

This paper introduces TIES, a dynamic token selection framework for vision-language-action models that improves efficiency and robustness by leveraging inter-layer token ranking consistency, outperforming static attention-based methods.

Contribution

TIES is a novel, adaptive token selection method that balances attention magnitude with ranking consistency, enhancing model performance without extra training.

Findings

Increases success rates by 6% on benchmark tasks.

Reduces token usage by 78%.

Generalizes well across different decoders and benchmarks.

Abstract

Vision-Language-Action (VLA) models excel in robotic manipulation but suffer from significant inference latency due to processing dense visual tokens. Existing token reduction methods predominantly rely on attention magnitude as a static selection. In this work, we challenge this assumption, revealing that high-attention tokens are task-dependent and can even degrade policy performance. To address this, we introduce \textbf{TIES} (\textbf{T}au-guided \textbf{I}nter-layer \textbf{E}fficient \textbf{S}election), a dynamic framework guided by inter-layer token ranking consistency. By adaptively balancing attention magnitude with ranking consistency, TIES ensures robust token selection without requiring additional training. On the CogACT + SIMPLER benchmark, TIES improves average success rates by 6\% while reducing token usage by 78\%, and demonstrate strong generalization across diverse decoders and benchmarks.