PaceVLA

PaceVLA is a unified framework that improves execution efficiency through an adaptive temporal execution mechanism while maintaining task success rate.

PaceVLA's action expert is a single shared backbone network dedicated to action generation that receives visual and language representations and outputs latent representations of action sequences, rather than multiple expert networks.

PaceVLA introduces a quantization-guided mode selection scheme that judges whether the current state is suitable for coarse execution, and uses a context-conditioned router to select the execution mode at each inference step based on visual observations, language instructions, and historical action context.

PaceVLA's router is essentially a lightweight classifier whose decisions are based on the current observation context rather than fixed rules; routing errors affect both speed and success rate.

PaceVLA was evaluated on three multi-task simulation benchmarks and six real-world tasks, covering multiple robot embodiments, including long-horizon tasks and dexterous manipulation.

Experimental results show that PaceVLA improves execution efficiency while maintaining task success rate; adaptive quantization does not trade reliability for speed.

The real-world tasks include precision-sensitive manipulation scenarios, a domain where fixed quantization tends to fail; PaceVLA's performance on these tasks validates the effectiveness of quantization-guided mode selection.

The paper abstract does not disclose the specific efficiency gains or success rate figures; readers need to consult the full text for detailed data.