Fluid-Agent Reinforcement Learning
MARL with dynamic agent populations.
Many multi-agent reinforcement learning settings assume a fixed number of agents. Fluid-agent RL studies settings where agents may create other agents, change the population size, or reorganize the team as part of the learning process.
This project develops formal models for fluid-agent games and evaluates MARL algorithms in environments where population control is part of the strategy, including fluid variants of predator-prey and level-based foraging.
Keywords: multi-agent reinforcement learning, population change, spawning, stochastic games, equilibrium reasoning.