PANDO：透過線上技能蒸餾實現高效多模態AI代理

摘要

近期多模態網頁代理的進展，往往依賴於增加推理階段的計算量，包括展開搜索、驗證器輪次、離線技能發現以及專家模型堆疊。這引發了一個核心問題：網頁代理能否隨著經驗累積變得更有效率，而非更昂貴？我們首先分析了VisualWebArena中的軌跡，並找出三個反覆出現的效率瓶頸：重複動作循環、隱藏發現成本，以及低提示快取重複利用率。接著，我們提出PANDO，這是一個一次展開的線上技能蒸餾框架，能維護結構化的技能庫，並結合進度反思、基於信心的技能降級、層次化路由、視覺壓縮，以及快取感知提示技術。在完整的910項VisualWebArena任務中，PANDO達成了58.3%的成功率，優於SGV（54.0%）與我們重現的WALT（45.2%），同時比SGV減少58%的token用量，比WALT減少61%，且無需任何預先評估的發現預算。一項包含300項任務的消融研究進一步顯示，規則與常規貢獻了大部分的成功提升，而路由、壓縮與快取感知提示則將更大的技能庫轉化為更低的邊際token成本。最後，我們引入了三項軌跡層級的效率指標——動作重複率、步驟開銷比，以及提示快取利用率——使效率在終端成功率之外也能被清晰觀察。

English

Recent advances in multimodal web agents often rely on increased inference-time computation, including rollout search, verifier passes, offline skill discovery, and specialist model stacks. This raises a central question: can a web agent become more efficient as it accumulates experience, rather than more expensive? We first analyze trajectories from VisualWebArena and identify three recurring sources of inefficiency: repeat-action loops, hidden discovery costs, and low prompt-cache reuse. We then introduce PANDO, a single-rollout online skill-distillation framework that maintains a structured Skill Library and combines progress reflection, confidence-based skill demotion, hierarchical routing, visual compression, and cache-aware prompting. On the full set of 910 VisualWebArena tasks, PANDO achieves a 58.3% success rate, outperforming SGV (54.0%) and our WALT reproduction (45.2%), while using 58% fewer tokens than SGV and 61% fewer tokens than WALT, without any pre-evaluation discovery budget. A 300-task ablation further shows that rules and routines provide most of the success gains, while routing, compression, and cache-aware prompting convert the larger skill library into lower marginal token cost. Finally, we introduce three trajectory-level efficiency metrics -- Action Repetition Rate, Step Overhead Ratio, and Prompt Cache Utilization -- to make efficiency visible beyond terminal success.