A new framework has been developed to learn admissible heuristics for optimal planning, addressing the challenges of overestimation in traditional methods. This machine-learned approach guarantees admissibility while improving efficiency in node expansions during planning tasks.