This repository includes the paper for the swarm robotics research I conducted at the University of Minnesota. The goal of this research was to develop a novel occupancy grid-based swarm foraging algorithm, simulate the algorithm on a swarm robotic system using the Webots simulator, and determine the relationship between sensor visibility and foraging efficiency.
Swarm robotics is the study of decentralized multi-robot systems that involve robots with simple, limited functionalities such as local sensing, low processing power, and limited inter-robot communication. These limitations introduce the need for effective indirect communication and mapping algorithms, especially for foraging missions, in which robots are tasked with finding and collecting randomly distributed objects in unfamiliar environments. Previous studies have examined several methods of indirect swarm communication and navigation, such as random walk models and artificial pheromone stigmergy. This study explores the viability of occupancy grid mapping as a method of indirect communication for a swarm foraging task. Occupancy grid mapping is a method of generating maps of an environment by representing the occupancy of each location in an environment using a probability array. With its robustness to sensor measurement uncertainty, occupancy grid mapping addresses the issue of generating maps in noisy environments. This study extends this technique to the multi-robot case by introducing a novel occupancy grid-based swarm foraging algorithm and determining a relationship between robot sensor visibility and foraging efficiency. Results of simulation testing in the Webots simulation software for a swarm of five robots under a fixed time constraint reveal that this algorithm yields a 69-81% average object collection rate in low-density object distributions and a 66-75% average object collection rate in high-density object distributions. Additionally, as the robots’ sensor noise levels are increased, object collection rates decrease only marginally, demonstrating the robustness of this algorithm to uncertainties in sensor measurements.