Plover: Steering GUI Agents through Plan-Centric Interaction
The paper’s central move is to make a GUI agent’s plan visible and editable while it works.
Plover uses a planner-executor setup so users can inspect execution, revise localized parts of a plan, add natural-language guidance, or intervene from screenshots. The authors say this lets repairs preserve prior progress instead of restarting the task. A six-person formative study shaped the interaction design, followed by tests on benchmark failure repair and workflow scenarios. Their results suggest many GUI-agent failures can be repaired when replanning is explicit and user corrections stay local. ArXiv · AI/CL/LG's note
Plover uses a planner-executor setup so users can inspect execution, revise localized parts of a plan, add natural-language guidance, or intervene from screenshots. The authors say this lets repairs preserve prior progress instead of restarting the task. A six-person formative study shaped the interaction design, followed by tests on benchmark failure repair and workflow scenarios. Their results suggest many GUI-agent failures can be repaired when replanning is explicit and user corrections stay local. ArXiv · AI/CL/LG's note
score 4