Universal Artificial Pheromone Framework with Deep Reinforcement Learning for Robotic Systems

Abstract

Pheromone-based communication has been adopted into robotic systems inspired by social insects as an alternative communication strategy, promising for dynamic and complex environments. For this reason, artificial pheromone communication system for robotic systems, especially for swarm robotic systems, has been proposed with diverse mediums such as light and virtual environment. However, the low versatility of each method makes it difficult to utilise the benefits of pheromone-based communication in diverse robotic platforms and environments. In this paper, we proposed PhERS (Pheromone for Every RobotS) framework designed to increase versatility, aiming to boost research and applications of pheromone-based communication in robotics. To validate the framework, we conducted experiments with simulated robots manoeuvred by hand-tuned controller performing navigation and collision avoidance tasks. As another contribution, we proposed a novel Deep Reinforcement Learning (DRL)-based controller for robots utilising pheromones to overcome the limitations of hand-tuned controller. Experiments and observed results demonstrated the feasibility of using the proposed framework in a robotic scenario, showing that DRLbased controller outperforms the baseline hand-tuned controller in a dynamic environment.