For these three years, we have been trying to build an accurate world models for the agent to select a best/better behavior. This required a lot of resources of the system, though it is still impossible to get the exact world model. So we stood back to the point: why are we trying to get an accurate world model. It is only because we believed that an accurate world model would help the agent determine the best/better behavior to win the soccer game. If the agent could select best/better behavior with less information, our requirement is fulfilled. So, this year we have worked to build a world model that has minimum but enough for the agent to determine its behavior.
展开▼