Multimodal Web Navigation with Instruction-Finetuned Foundation Models

arXiv:2305.11854v3 Announce Type: replace
Abstract: The progress of autonomous web navigation has been hindered by the dependence on billions of exploratory interactions via online reinforcement learning, and domain-specific model designs that make it difficult to leverage generalization from rich out-of-domain data. In this work, we study data-driven offline training for web agents with vision-language foundation models. We propose an instruction-following multimodal agent, WebGUM, that observes both webpage screenshots and HTML pages and outputs web navigation actions, such as click and …

abstract agents arxiv autonomous cs.ai cs.lg data data-driven designs domain exploratory foundation interactions multimodal navigation offline online reinforcement learning progress reinforcement reinforcement learning stat.ml study training type via web web navigation work