In this work, a novel value function-based reinforcement learning (RL) approach, descending dynamic policy programming (DDPP) is proposed to address the issues of sample-efficiency and learning ...