Search papers, labs, and topics across Lattice.
The paper introduces VLA-InfoEntropy, a training-free method to accelerate inference in Vision-Language-Action (VLA) models by dynamically focusing on informative visual regions and semantically relevant text tokens. It uses image entropy to identify texture-rich visual regions and attention entropy to pinpoint semantically relevant text tokens, combined with timestep information for dynamic transition. Experiments demonstrate that VLA-InfoEntropy reduces inference parameters and accelerates inference speed compared to existing approaches.
Achieve faster VLA inference without retraining by using image and attention entropy to dynamically focus on the most relevant visual and textual information.
Vision-Language-Action (VLA) models integrate visual perception, language understanding, and action decision-making for cross-modal semantic alignment, exhibiting broad application potential. However, the joint processing of high-dimensional visual features, complex linguistic inputs, and continuous action sequences incurs significant computational overhead and low inference efficiency, thereby hindering real-time deployment and reliability. To address this issue, we use image entropy to quantify the grayscale distribution characteristics of each visual token and introduce attention entropy to capture the distribution of attention scores over task-related text. Visual entropy identifies texture-rich or structurally informative regions, while attention entropy pinpoints semantically relevant tokens. Combined with timestep information, these metrics enable a dynamic transition strategy that shifts the model's focus from global visual features to attention-guided local informative regions. Thus, the resulting VLA-InfoEntropy method integrates spatial, semantic, and temporal cues to reduce redundancy while preserving critical content. Extensive experiments show that our method reduces inference parameters, accelerates inference speed, and outperforms existing approaches.
