Real-time landmark modelling for visual-guided walking robots

The environment sensing through visual perception can be obtained by an efficient and reliable landmark modelling. This paper proposes a novel landmark modelling approach to the classification of both the predefined and novelty features in the robot environment. The approach integrates image descriptors for defined landmarks (natural and artificial) and novelty features being dynamically detected. It has been implemented in a quadruped walking robot that has a single camera in its head. Some experimental results are presented to demonstrate the feasibility and performance of the proposed landmark models.