On the other hand, when these agents are trained on the main game, they are unable to make significant progress. While playing the minigames, these agents gain the ability to perform at a level of play that is equivalent to that of a beginner. Finally, we present some preliminary results that serve as a baseline for canonical deep reinforcement learning agents when applied to the StarCraft II domain.
We provide some preliminary results as a baseline for neural networks that have been trained to predict game outcomes and player actions using this data. This data comes from playing the game multiple times. We also provide an accompanying dataset of game replay data from human experts for the primary game maps. In addition to the main game maps, we also provide a collection of minigames that each focus on a particular aspect of the gameplay in StarCraft II. We provide an open-source Python-based interface for communicating with the game engine and describe the observation, action, and reward specification for the StarCraft II domain.
It is a multi-agent problem with multiple players interacting with one another there is imperfect information as a result of a partially observed map it has a large action space that involves the selection and control of hundreds of units it has a large state space that must be observed solely from raw input feature planes and it has delayed credit assignment, which requires long-term strategies that extend over thousands of steps. This domain presents a new and significant challenge for reinforcement learning because it represents a class of problems that is more difficult than those that have been considered in the majority of previous work. This paper presents SC2LE 5120x1440p 329 starcraft ii image, an online environment for reinforcement learning that is modelled after the video game StarCraft II. A New Obstacle to Conquer for Reinforcement Learning in StarCraft II
0 kommentar(er)
