A New Approach in Agent Path-Finding using State Mark Gradients

• Main Authors: Florin Leon, Mihai Horia Zaharia, Dan Galea
• Format: Article
• Language: English
• Published: Institute of Mathematics and Computer Science of the Academy of Sciences of Moldova 2005-01-01
• Series: Computer Science Journal of Moldova
• Subjects: Artificial intelligence; path-finding; maze; reinforcement learning; Q-learning; LRTA*; agents
• Online Access: http://www.math.md/files/csjm/v12-n3/v12-n3-(pp406-423).pdf

Since searching is one of the most important problem-solving methods, especially in Artificial Intelligence where it is often difficult to devise straightforward solutions, it has been given continuous attention by researchers. In this paper a new algorithm for agent path-finding is presented. Our approach is based on environment marking during exploration. We tested the performances of Q-learning and Learning Real-Time A* algorithm for three proposed mazes and then a comparison was made between our algorithm, two variants of Q-learning and LRTA* algorithm.