Estimating the dominant frequencies of real-valued cyclic processes in natural environments for long-term operation of autonomous robots

Abstract

Autonomous mobile robot localisation, planning, and navigation methods typically rely on environmental representations. Previous research has shown that the temporal dynamics captured by specialised models help autonomous robots to operate for longer periods with better efficiency. One of the leading approaches uses maps enhanced by frequency analysis to model and predict repeating cycles of activity in the environment. However, this approach implicitly relies on prior knowledge of the expected periodicities, such as days and weeks, encoded by the human designers of the system. This paper presents a new method to automatically search the robot's observations for dominant frequencies, leading to a more general method than the previous approach for frequency map enhancement. The proposed algorithm extends the problem definition from binary observations also to real-valued time series, making it applicable to a broader spectrum of robotic tasks. We show that the new method can be implemented in robotic systems operating without prior knowledge of the underlying processes that influence the dynamics of the working environment across a wide variety of tasks similar to the long-standing state-of-the-art. We hypothesise that an autonomous robot using the proposed improvement can be deployed to unprecedented environments.