Date of Award
Master of Science in Systems Engineering
Mechanical, Industrial and Systems Engineering
This work examines several approaches to using multiple robotic search agents to locate a hidden object within a bounded space. Animal swarm behavior is used as model for robots using collaborative tactics with limited intelligence and sensing capabilities. Leveraging group behavior creates opportunities for the robotic swarm to perform better than an equal number of individual robots. Investigated advantages of using a robotic swarm include added endurance, shared GPS accuracy, robustness against evasive maneuvers, and (most importantly) decreased search time.
An object-oriented program written in Python is used to compare eight group-search approaches. A simulation counts the number of steps each group must take before at least one agent in the group stumbles upon the hidden object, which is located in a random set of coordinates. The program is a prototype with alterable parameters to fit various environments, and useful for testing of future strategies.
Simulation-based results align with the theory that sharing sensor data can result in longer battery life, and decrease search time as a result of fewer refuel trips. A differential-GPS sharing method is also used successfully to reduce GPS error and improve search efficiency. A memory-based method which marks positions to reduce duplicate visits has less conclusive results, improving some simulations but leading to an overall decrease in search efficiency.
The final phase of the experiment includes a dynamic aspect in which the hidden object attempts to evade the search agents. The results of the experiment show that having the swarm organize into wide shapes that move as a unit makes it more difficult for the object to get away. This provides incentive for implementation of self-organization algorithms for group robotic searches.
Maio, Justin P., "OPTIMIZING AUTONOMOUS MULTI-AGENT SEARCH PATTERNS USING SWARM INTELLIGENCE" (2023). Open Access Master's Theses. Paper 2326.
Available for download on Wednesday, May 08, 2024