Path Navigation For Indoor Robot With Q-Learning

AbstractA Q-learning based path navigation method is proposed and validated in this paper for solving the moving control along specified path of real indoor mobile robot. A grid and topological indoor corridor environment map is employed and characterized by a set of geometrical scale invariant key-points. The navigation strategy is composed of on-line and off-line stages with the same components redefinitions or definitions of Q-learning. During the off-line learning stage, the personal computer records the optimal path after computer learning simulation, and then the path is sent to the robot through wireless data radio with RS232. At the on-line navigation stage, the robot calculates the relative positions between the locations along this optimal path, and then navigates the environment autonomously. The experiments on computer simulation and an actual robot have been verified the superior effectiveness and applicability of the proposed strategy.