Short- and long-term adaptation of visual place memories for mobile robots

This paper presents a robotic implementation of a human-inspired memory model for long-term adaptation of spatial maps for navigation in changing environments. The robot uses an appearance-based representation of its workplace as a map, where the current view and the map are used to estimate the robots current position in the environment. Due to the nature of real-world environments such as houses and offices, where the appearance keeps changing, the map may become out of date after some time. To solve this problem the robot needs to adapt the map continually in response to the changing appearance of the environment. In this work we use local features extracted from panoramic images to represent the appearance of the environment. Adopting concepts of short-term and long-term memory, our method updates the group of feature points for the image representation of a particular place. Experiments using robot sensor data collected over a period of 2 months show that the implemented model is able to adapt successfully to changes