LLM-generated robot policy code is unreliable, with failures clustering into four behavior types that grow with task complexity and shrink with instruction detail; a failure-feedback retry improves success up to 35%.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.RO 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
RoboInspector: Unveiling the Unreliability of Policy Code for LLM-enabled Robotic Manipulation
LLM-generated robot policy code is unreliable, with failures clustering into four behavior types that grow with task complexity and shrink with instruction detail; a failure-feedback retry improves success up to 35%.