Authors: Peter K. K. Loh, Rahul Bhasker, Edmond C. Prakash
Abstraction Moving Target Search (MTS) was shown to exhibit competitive performance against other existing MTS algorithms even with large problem spaces . However, the original design of the Abstraction MTS algorithm does not incorporate learning. Hence, its behaviour is deterministic and its performance is constant regardless of the number of times the simulation is repeated over the same problem space. In this paper, we investigate the incorporation of heuristic-based learning into Abstraction MTS to enhance its adaptiveness. In our new algorithm, Heuristic Abstraction MTS, we attempt to minimize the computational as well as information overheads that may be introduced when
incorporating heuristic-based learning into the design. We conduct performance simulations with an agent and a moving target within randomly generated mazes of increasing sizes and reveal that the Heuristic Abstraction MTS outperforms the original Abstraction MTS.