Cleaning Agents with DeepQLearning – A perfomance comparison betwen MLP and RNN networks

In this report, I investigate the application of Deep Reinforcement Learning (DRL) techniques, tailored for Multilayer Perceptron (MLP) and Long ShortTerm Memory Recurrent Neural Networks (LSTM RNN), within VMAS (Vectorized Multi-Agent Simulator). My study introduces the "Cleaning Agents" scenario, where N agents are tasked with efficiently cleaning M stationary targets in a 2D space. The central goal of this project is to evaluate and compare the impact of DRL methods, specifically MLP and LSTM RNN, on agent coordination and decision-making within this scenario. To achieve this goal, I employ time-based performance metrics to assess the efficiency and effectiveness of these DRL variants. By quantifying the temporal aspects of agent behavior, I aim to provide practical insights into the application of DRL in multi-agent systems. This comparative analysis offers valuable guidance for selecting the most suitable DRL architecture for similar multi-agent tasks.