Pose estimation of a mobile robot based on network sensors adaptive sampling

Abstract

This paper covers the use of an adaptive sampling strategy on a sensor network for the estimation of the pose of a remotely controlled mobile robot. The more measurements can be taken, the better the estimation becomes, but if some level of uncertainty is tolerable, the amount of measurements can be significantly reduced. The proposed adaptive sampling technique is triggered by the estimation error covariance matrix, computed by Kalman filter algorithms. This sampling strategy leads to the minimum amount of measurements needed to achieve a certain estimation accuracy, reducing the load on important resources such as computational power and network communications. An example of use is given as the pose estimation of a wheeled robot remotely controlled. A network of camera sensors provides pose measurements allowing the application of an Unscented Kalman Filter estimator in the remote centre.