University of New Hampshire
Mentor: Dr. Wheeler Ruml, UNH Department of Computer Science
Empirical Analysis of Real-Time Search Algorithms in a Dynamic, Multi-Agent Domain
An agent’s ability to plan autonomously in order to achieve its goals is necessary if we want it to achieve goals without a human giving it step-by-step instructions. Real-time search algorithms allow an agent to plan its next action and choose that action within a pre-specified time limit. One example is robot motion planning. In a planned environment, there may be people moving around that the robot must avoid hitting. After each planning episode ends, the robot takes the selected action and reevaluates the surrounding area. However, real-time search algorithms have not been tested for robot motion planning in a domain that features moving obstacles and requires real-time response. My project will determine if real-time search algorithms become impractical when used for robot motion planning in a dynamic, multi-agent domain and identify the causes of their inadequacy. These results will provide a foundation for creating a new, more efficient algorithm.