Search papers, labs, and topics across Lattice.
This paper identifies the latency in autoregressive decoding as a critical factor causing GUI agents to miss transient events, leading to delayed but correct actions. To address this, the authors introduce Adaptive Anticipatory Policy Trees (AAPT), which utilize a pre-compiled policy tree to enable immediate action execution during decision-making without generating new text. Experimental results show that AAPT significantly improves success rates from 0.50 to 0.79 in contested decision windows while maintaining zero incorrect actions, outperforming traditional reactive methods in scenarios where candidate actions can be anticipated.
AAPT boosts GUI agent success rates by 58% in decision-critical moments, eliminating delays without sacrificing accuracy.
Computer-use agents often fail on transient GUI events because they produce the correct action only after the relevant window has already closed. We identify the main cause as expensive autoregressive decoding on the decision-time critical path. We propose Adaptive Anticipatory Policy Trees (AAPT), which eliminates this delay without modifying the underlying model. During idle screen periods, the same frozen multimodal model constructs a bounded conditional policy tree with observable guards, pre-authorized actions, and branch-specific deadlines. The tree is sized to cover the model's own decoding latency. When an event occurs, a lightweight observer matches change-gated frames to a prepared branch and immediately executes the corresponding action without generating new text. In paired trials with pre-registered endpoints and exact McNemar tests, AAPT improves the success rate from 0.50 to 0.79 within a contested decision window ($p=1.8\times10^{-3}$), while producing no incorrect actions. Both open-loop and predict-and-replan baselines achieve zero success because they still decode during execution. A preparation-time sweep shows that the gain emerges where the latency-based tree-sizing rule predicts, and ablations reveal three key requirements: fast observer decoding, valid tree planning, and accurate branch routing. A pre-registered oracle probe rejects our initial hypothesis and instead points to branch routing as the causal bottleneck. We further reproduce the effect on an independent general-purpose multimodal model over 126 paired trials ($p=4.9\times10^{-13}$). On an external benchmark, AAPT matches the overall performance of a reactive baseline, although the two methods exhibit complementary strengths. Together, these results suggest that AAPT performs best when candidate actions can be enumerated in advance, whereas reactive execution remains stronger when they cannot.
