A closed-loop GUI grounder that reflects on its own predicted coordinates via marker-rendered patches and a locate/confirm protocol, trained with SFT then GRPO, achieves state-of-the-art grounding and refusal accuracy.
Proceedings of the AAAI Conference on Artificial Intelligence , volume=
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
LookAgain: Closed-Loop GUI Grounding with Visually Grounded Reflection
A closed-loop GUI grounder that reflects on its own predicted coordinates via marker-rendered patches and a locate/confirm protocol, trained with SFT then GRPO, achieves state-of-the-art grounding and refusal accuracy.