Self localisation using embodied data for a hybrid leg-wheel robot

Robotic systems that are able to navigate autonomously in unstructured outdoor terrain have a large potential in a number of applications like planetary exploration or search and rescue scenarios. Localisation is usually performed through dead-reckoning with the help of visual means. The approach described in this paper uses only sensory information internal to the system to localize in a partially known environment. A model of the robot and sensory information on its configuration and orientation are used to match candidate contact points with an environment model. A particle filter implementation is developed, which uses this information together with the odometry to track the pose of the robot. The experiments conducted on a hybrid leg-wheel robot show that the approach is able to track the position of the robot within an average error of 0.5 m for test runs of up to 140 m distance travelled. One potential of this approach is to reduce the requirements on the visual parts when integrated into SLAM frameworks.