Learning of the Exploratory Behavior of Mobile Robots for Environmental Segmentation

A model that can describe the operating environment of robot is necessary for mobile robots to operate effectively in real-world environments. Environmental segmentation, which classifies environments based on their features, is useful for environmental modeling. To realize autonomous environmental segmentation by a robot, the robot must explore through its environment and collect data. In previous studies, the exploratory behavior of a robot for data collection was predesigned by a designer. In addition to the information on environment, those on physical features of robots are essential for design the robot's behavior. Additionally, the exploratory behavior must be redesigned based on the changes in the environment. In this study, an autonomous learning method is proposed for the exploratory behavior of mobile robots using reinforcement learning. Moreover, the effectiveness of the proposed method is verified via a simulation experiment. Results revealed that the feature-based environmental segmentation can be performed using the sensor information obtained from the exploratory behavior of robot, realized with the proposed method. By using the proposed method, robots can be expected that automatically explore new environments "collect data" and segment the environment.