A two-layer navigation architecture for autonomous indoor exploration is proposed. As doors in the indoor environments are ubiquitous and informative for navigation, the high-level layer uses a stereo vision based algorithm to detect doors in order to generate a series of goal points. The low-level layer makes the robot avoid obstacles and navigate to the goal points by an improved dynamic window approach (DWA). This approach can solve the local-minima problem of DWA and can obtain smooth motion controls. The combination of high-level door-guidance and low-level goal-directed navigation enables the robot to make a door-to-door exploration. The proposed architecture was implemented on a Pioneer3 robot. Experiments show that the door detection algorithm and the improved DWA work reliably under various environments, and the robot can efficiently fulfill the task of autonomous indoor exploration.
展开▼