Deep RL Bootcamp Lecture 9 Model-based Reinforcement
So, the process is similar to one-to-many RNN?
learn much more efficiently than model-free method
iteratively get better
less than 300 trials ~ 25min robot time per task
visual prediction from the observation
during train of model, there is no reward. Some random motions are programmed. at the task time, there is a reward function, basically trying to move a pixel to the goal position.