Embodied LLMs achieve higher puzzle-solving success with raw RGB observations than ground-truth symbolic ones, with moderate action-outcome noise boosting rates 2.85-fold by reducing repetitive loops.
Frontiers in Behavioral Neuroscience17, 1230082 (2023)
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AI 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Probing Embodied LLMs: When Higher Observation Fidelity Hurts Problem Solving
Embodied LLMs achieve higher puzzle-solving success with raw RGB observations than ground-truth symbolic ones, with moderate action-outcome noise boosting rates 2.85-fold by reducing repetitive loops.