An Adaptive Memory Model for Long-Term Navigation of Autonomous Mobile Robots

Abstract

This paper introduces an environmental representation for autonomous mobile robots that continuously adapts over time. The presented approach is inspired by human memory information processing and stores the current as well as past knowledge of the environment. In this paper, the memory model is applied to time-variant information about obstacles and driveable routes in the workspace of the autonomous robot and used for solving the navigation cycle of the robot. This includes localization and path planning as well as vehicle control. The presented approach is evaluated in a real-world experiment within changing indoor environment. The results show that the environmental representation is stable, improves its quality over time, and adapts to changes.