Adaptive learning for robots in public spaces

Abstract

Proper functioning of robots deployed in public spaces often require extensive knowledge of its environment of use, which is completely unknown prior to deployment. The methods for acquiring and utilizing such knowledge also varies depending on the nature of the public space and the tasks the robot needs to perform. This calls for development and application of adaptive learning methods specifically designed to take into consideration the nature and key properties of various public spaces and robotic tasks. In this paper, we study typical types of public spaces for deployment of robots, and analyze robotic tasks required in each type of space to derive common capabilities that the robots need to have. We then consider three adaptive learning methods: (1) autonomous learning, (2) unsupervised learning from real-time on-site data, and (3) guided learning. Applicability of the methods to improve each common capability and possible means of application are further discussed.