Reset-free guided policy search: Efficient deep reinforcement learning with stochastic initial states | IEEE Conference Publication | IEEE Xplore