Swarm Pathfinding Algorithms

Project Summary

This project focuses on swarm robotics and its application to multi-agent path planning in unfamiliar environments with unknown static obstacles. Three emerging algorithms are evaluated: Multi-Agent Deep Deterministic Policy Gradient (MADPG). Hybrid Simplified Grey Wolf Optimization with Modified Symbiotic Organism Search (HSGWO-MSOS), and Improved Artificial Potential Field (APF). The A* Search algorithm is used as a baseline algorithm for comparison with experimental methods. We compared path planning of these algorithms in a 2D simulated environment, considering execution time, completeness, scalability, and efficiency. In order to test and evaluate these algorithms, we created a swarm simulation software using python's pygame package. The simulation software simulates a number of agents navigating an environment to a known goal, through unknown obstacles.

I worked on a team of 4, and was the leading software contributor.

Data Analysis

I used batch scripting in order to generate data from hundreds of simulations for each algorithm. Data such as runtime, pathlength, swarm size, completion percentage, and obstacle difficulty were collected. We then exported data to excel and used it to compare the 3 algorithms in our report.