An extended policy gradient algorithm for robot task learning | IEEE Conference Publication | IEEE Xplore