Plover: Inspectable Plans for GUI Agent Repair
Plover makes GUI-agent plans persistent, visible, and editable so users can inspect execution, guide replanning, and repair failures without discarding prior progress.
TL;DR — Plover is a plan-centric, vision-based GUI automation system that externalizes task plans and replanning as persistent, inspectable, and revisable artifacts. The abstract claims this enables explicit supervision, localized correction, natural-language guidance, screenshot-grounded interventions, and repair that preserves prior progress.
Core idea
Plover addresses a weakness in autonomous GUI agents: they may operate over screenshots and natural-language instructions, but their plans and adaptation processes can remain hidden. When interfaces change, the agent can drift from the user's intent and be hard to inspect or correct.
Plover's central idea is to externalize planning. Task plans and replanning are treated as persistent, inspectable, and revisable artifacts rather than hidden internal state.
The abstract describes an architectural and interaction contribution, not a new benchmark result with reported numeric gains. Its focus is on making GUI automation more transparent, controllable, adaptable, and repairable.
Full analysis, extracted claims, numerical results, entity graph, FAQ, related work, applications, and BibTeX are available via x402 micropayment.