The article presents the concept of uniform action costs in automated planning. It suggests that agents may prefer plans where action costs are as uniform as possible, providing stability and predictability. This is particularly relevant when humans are the agents executing the plans. The paper adapts three uniformity metrics to automated planning and introduces new methods to optimize the sum of action costs and their uniformity. The concept is demonstrated through examples like network routing and navigation scenarios, where uniform action costs can reduce bottlenecks and provide better service. The paper also discusses the relevance of this concept in financial planning.
Publication date: 16 Feb 2024
Project Page: https://www.jpmorgan.com/ai-research
Paper: https://arxiv.org/pdf/2402.09877
