LWM-Planner improves LLM agent performance by extracting atomic facts from past episodes and using them to condition a recursive lookahead search with an LLM-based world model.
• Qualitative Goal: The LLM engages in a form of abductive reasoning to hypothesize underlying environmental properties or dynamics
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Fact-Augmented Lookahead Planning for LLM Agents
LWM-Planner improves LLM agent performance by extracting atomic facts from past episodes and using them to condition a recursive lookahead search with an LLM-based world model.