Event

PaceVLA proposes an adaptive temporal action quantization framework for VLA

1 reports · 1 sources

Hojin Jeon and other researchers propose the PaceVLA framework, which improves the execution efficiency of vision-language-action models (VLA) through adaptive temporal action quantization. The framework builds multi-granularity action decoders on a shared action expert and uses quantization-feasibility-guided mode selection and context-conditioned routing to decide at each inference step whether to adopt coarse-grained execution. The paper states that the experiments cover three multi-task simulation benchmarks and six real-world tasks, and that the framework improves execution efficiency on multiple embodiments (including long-horizon and dexterous manipulation) while maintaining task success rates.