Provably Complete Generalized Planning with LLMs
This work introduces a semantic-preserving PDDL-to-Lean conversion, and uses an LLM to generate both the generalized plan and the formal proof that it solves every instance satisfying the domain constraints, and evaluates this approach on 13 commonly used benchmark domains.
Katharina Stein, Chaahat Jain, J. Hoffmann et al.
· 0 citations