Local Robot Navigation Based on an Active Visual Short-Term Memory

Vision devices are today one of the most often used sensory elements in autonomous robots. Some of their hindrances are the difficulty in extracting useful information from the captured images and the small visual field of regular cameras. Visual attention systems and active vision may help to overcome them. This work proposes a dynamic visual memory to store the information gathered from a continuously moving camera onboard the robot and an attention system to choose where to look at with such mobile camera. The visual memory is a collection of relevant task-oriented objects and 3D segments, and its scope is wider than instantaneous field of view of the camera. The attention system takes into account the need to reobserve objects in the visual memory, explore new areas and test hypothesis about object existence in the robot surroundings. The system has been programmed and validated in a real Pioneer robot that uses the information in the visual memory for navigation tasks.