Dynamic Influence Based Fitness Shaping For Cooperative Coevolution

Multiagent settings are naturally characterized by coevolutionary dynamics, where agents must adapt and learn in the context of their teammates. A key challenge in such domains is determining how to credit an agent for its individual contribution to team performance. Fitness shaping approaches partially address this by identifying and isolating an agent's direct contribution to team success. However, when an agent's contribution is indirect, such as influencing other teammates to succeed, existing methods fail to capture its impact on the team. This paper presents a comprehensive mathematical and algorithmic formulation of Dynamic Influence Based Fitness Shaping, a fitness shaping method for heterogeneous teams that isolates both direct and indirect contributions by evaluating how agents influence others over time. By considering inter-agent influence at a high temporal resolution, Dynamic Influence extracts direct credit from indirect interactions. Results in an autonomous aerial and terrestrial vehicle coordination problem demonstrate 30% performance improvement in a team with 32 vehicles and 12% improvement in a sequence-based variant with 10 vehicles. The contributions of this paper are a mathematical framework that unifies Dynamic Influence with prior shaping approaches, a detailed breakdown of the Dynamic Influence algorithm, and a supporting coevolutionary framework that extends Dynamic Influence to coordinate many agents across sequence-based tasks.