Agent Motion Planning with GAs Enhanced by Memory Models

摘要

The Tartarus problem may be considered a benchmark problem in the field of robotics. A robotic agent is required to move a number of blocks to the edge of an environment. The location of the blocks and position of the robot is unknown initially. The authors present a framework that allows the agent to learn about its environment and plan ahead using a GA to solve the problem. The authors prove that the GA based method provides the best published result on the Tartarus problem. An exhaustive search is used within the framework as a comparison, this provides a higher score still. This paper presents the two best Tartarus results yet published