Abstract

Recent Vision-Language-Action (VLA) models built on pre-trained Vision-Language Models (VLMs) require extensive post-training, resulting in high computational overhead that limits scalability and deployment. Existing sparsification strategies—such as Mixture-of-Depths, layer skipping, and early exit—fall short by neglecting the semantic coupling across vision-language-action modalities, and focusing narrowly on intra-LLM computation while overlooking end-to-end coherence from perception to control. To address these challenges, we propose CogVLA, a Cognition-Aligned Vision-Language-Action framework that leverages instruction-driven routing and sparsification to improve both efficiency and performance. CogVLA draws inspiration from human multimodal coordination and introduces a 3-stage progressive architecture. 1) Encoder-FiLM based Aggregation Routing (EFA-Routing) injects instruction information into the vision encoder to selectively aggregate and compress dual-stream visual tokens, forming a instruction-aware latent representation. 2) Building upon this compact visual encoding, LLM-FiLM based Pruning Routing (LFP-Routing) introduces action intent into the language model by pruning instruction-irrelevant visually grounded tokens, thereby achieving token-level sparsity. 3) To ensure that compressed perception inputs can still support accurate and coherent action generation, we introduce V‑L‑A Coupled Attention (CAtten), which combines causal vision-language attention with bidirectional action parallel decoding. Extensive experiments on the LIBERO benchmark and real-world robotic tasks demonstrate that CogVLA achieves state-of-the-art performance with success rates of 97.4% and 70.0%, respectively, while reducing training costs by 2.5× and decreasing inference latency by 2.8× compared to OpenVLA. CogVLA is open-sourced and publicly available at https://github.com/JiuTian-VL/CogVLA.

Overview Framework of CogVLA

CogVLA employs a cognition-aligned, instruction-driven routing & sparsification strategy for efficient action chunk prediction. Inspired by human multimodal coordination, it integrates task-guided visual aggregation, semantic pruning, and coherent decoding, ensuring efficient cross-modal representation alignment from perception to control.

Experiment

Table 1 and 2: Main Results of CogVLA on LIBERO and ALOHA.

Table 3: Speed improvement of CogVLA on LIBERO.

Qualitative Results

Visualization of the attention map of aggregation tokens. We visualize the attention maps generated by the cross-modal attention modules. As shown in the figure, the attention weights highlight task-relevant regions in the input image. These visualizations demonstrate that CogVLA's instruction-aware routing mechanisms effectively guide the perception module to attend to semantically meaningful areas, enabling robust visual grounding even in cluttered or ambiguous scenes.

Videos of Real-world Tasks

GALAXEA Platform

Instruction: Open the drawer,place the toy into the drawer, and then close it.

ALOHA Platform

Instruction: Open the drawer,place the toy into the drawer, and then close it.

Instruction: Fold the T-shirt.

Conclusion

We presented CogVLA, a cognition-aligned and instruction-driven Vision-Language-Action framework designed to address the computational inefficiencies and semantic fragmentation in existing VLA models. By integrating EFA-Routing, LFP-Routing, and CAtten into a unified 3-stage progressive design, CogVLA achieves effective vision sparsification and coherent cross-modal reasoning. Extensive evaluations on both the LIBERO benchmark and real-world robotic tasks demonstrate that CogVLA not only achieves state-of-the-art performance but also significantly reduces computational cost and inference latency. This work highlights the importance of instruction-driven multimodal sparsification in building scalable and efficient embodied AI systems.

BibTeX

@article{li2025cogvla,
      title={CogVLA: Cognition-Aligned Vision-Language-Action Model via Instruction-Driven Routing & Sparsification},
      author={Wei Li and Renshan Zhang and Rui Shao and Jie He and Liqiang Nie},
      journal={arXiv preprint arXiv:2508.21046},
      year={2025},
}