Learning to Adapt: Self-Improving Web Agent via Cognitive-Aware Exploration

In three linesSCALE is a self-improving framework for web agents using MLLMs. It employs three adversarial roles (Selector, Predictor, Judger) to autonomously explore agent limitations and expand cognitive boundaries. SCALE-Hop optimizes global planning via graph exploration. A SCALE-20k dataset from 19 real websites with 20k structured demonstrations validates the approach across multiple MLLMs.

Summary generated by Claude — human-verified