AutoRPA: Efficient GUI Automation through LLM-Driven Code Synthesis from Interactions
AutoRPA, introduced in a new arXiv paper (2605.21082v1), addresses the inefficiency of LLM-based GUI agents that repeatedly invoke reasoning for repetitive tasks. The framework automatically converts ReAct-style agent decision logic into robust RPA functions. It features a translator-builder pipeline: a translator agent converts hard-coded ReAct actions into soft-coded procedures, and a builder agent synthesizes RPA functions via retrieval-augmented generation over multiple trajectories. A hybrid repair strategy combines RPA execution with ReAct-based fallback for iterative refinement during code verification. Experiments across multiple GUI environments show that AutoRPA-generated RPA functions achieve runtime efficiency comparable to traditional RPA while reducing manual effort. The approach bridges the gap between flexible LLM agents and efficient RPA, enabling practical automation of repetitive GUI tasks without sacrificing robustness.
Reduces cost and effort for automating repetitive GUI tasks by replacing LLM calls with efficient RPA.