I am trying to construct an RL algorithm for managing a fleet of vehicles to maximize profit. As far as I know, the Sequential Decision Process can be decomposed into the following pic: My current method is based on value function approximation of post-decision States $S_t^x$ . But without the technique of Target Network and Experience Replay of DQN, my model is easily trapped at the local minima and can not take advantage of the earlier...
Originally asked by PokeLu on Stack Exchange (ai) (CC BY-SA 4.0).
