RA*pex: Efficient Approximate Multi-Objective Search Under Rulebooks

A recent study published on arXiv (2608.04398) presents RA*pex, a best-first search algorithm designed for multi-objective robotic planning guided by rulebooks. These rulebooks define the priority relationships between objectives such as safety, efficiency, and adherence to regulations, extending the concepts of Pareto and lexicographic dominance. Due to the high computational cost of determining all rulebook-optimal solutions, the authors introduce epsilon-rule-dominance, an approximate dominance concept, along with RA*pex to efficiently derive a compact set of epsilon-approximate rulebook-optimal solutions. The algorithm utilizes dimensionality reduction while honoring rule hierarchies, maintaining distinct closed sets, and conducting dominance evaluations on truncated and residual rule sets. Although the paper offers a formal analysis of RA*pex, not all details are included in the abstract. This research tackles the computational difficulties associated with robotic planning that involves multiple objectives and intricate priorities.