Learning to Plan via a Multi-step Policy Regression Method

We propose a new approach to increase inference performance in environments that require specific sequence of actions order be solved. This is for example the case maze where ideally an optimal path determined. Instead learning policy single step, we want learn can predict n advance. Our proposed method called horizon regression (PHR) uses knowledge environment sampled by A2C dimensional vector distillation setup which yields sequential per observation. test our on MiniGrid and Pong show drastic speedup during time successfully predicting sequences