Technical Report No. 265 - Abstract

Daniel Meyer-Delius, Maximilian Beinhofer, Wolfram Burgard:
Grid-Based Models for Dynamic Environments

The majority of existing approaches to mobile robot mapping assume that the world is static, an assumption which does not hold in most practical application domains. In this paper we present a probabilistic grid-based approach for modeling dynamic environments representing both, the occupancy and the dynamics of the corresponding area. We describe the environment as a spatial grid and use a hidden Markov model to represent the occupancy state and state transition probabilities of each grid cell. Our approach updates the occupancy state as observations become available. We describe an offline and an online technique to estimate the transition probabilities of the model from observed data. Experimental results show that our model is better suited for representing dynamic environments than standard occupancy grids. Furthermore, the results show that the explicit representation of the environment dynamics can be used to improve robot navigation.