Self-Imitation Learning for Robot Tasks with Sparse and Delayed Rewards | IEEE Conference Publication | IEEE Xplore