A Centralized PPO Agent-based Particle Swarm Optimization for Multi-UAV Source-Seeking

Kartik Vij¹    Arya Topale¹    Harikumar Kandath¹    J. Senthilnath²   

¹ Robotics Research Centre (RRC), IIIT Hyderabad, Hyderabad, India    ² Institute for Infocomm Research, A*STAR, Singapore National University of Singapore, Singapore   

This paper presents a centralized Proximal Policy Optimization (PPO) agent-based framework for Particle Swarm Optimization (PSO) based source-seeking for Uncrewed Aerial Vehicles (UAVs). The source is considered to be emitting a signal from an unknown location. The signal strength measured by individual UAVs drives the multi-UAV search in a Deep Reinforcement Learning (DRL) framework. We consider different variants of PSO like Acceleration-based PSO (APSO), Standard PSO (SPSO), and Adaptive Robotic PSO (ARPSO) that are commonly used in source-seeking problems. The coefficients of these PSO variants are selected by a PPO agent. The performance of the proposed method is demonstrated through extensive simulations highlighting time efficiency, sampling efficiency, and scalability with respect to the number of UAVs.