![]() These games offer multiple advantages: 1. They have clear objective measures of success 2. Computer games typically output rich streams of observational data, which are ideal inputs for deep networks 3. They are externally defined to be difficult and interesting for a human to play. Fourth, games typically last for many thousands of frames and actions, and the player must make early decisions (such as which units to build) with consequences that may not be seen until much later in the game (when the players’ armies meet), leading to a rich set of challenges in temporal credit assignment and exploration.Ĭomputer games provide a compelling solution to the issue of evaluating and comparing different learning and planning approaches on standardised tasks, and is an important source of challenges for research in artificial intelligence (AI). Furthermore, the set of legal actions varies as the player progresses through a tree of possible technologies. There are many different unit and building types, each with unique local actions. ![]() The player selects actions among a combinatorial space of approximately 10 8 possibilities (depending on the game resolution), using a point-and-click interface. Third, the action space is vast and diverse. Furthermore, there is a “fog-of-war”, obscuring the unvisited regions of the map, and it is necessary to actively explore the map in order to determine the opponent’s state. The map is only partially observed via a local camera, which must be actively moved in order for the player to integrate information. Second, it is an imperfect information game. It is also multi-agent at a lower-level: each player controls hundreds of units, which need to collaborate to achieve a common goal. First, it is a multi-agent problem in which several players compete for influence and resources. Thus, SC2LE offers a new and challenging environment for exploring deep reinforcement learning algorithms and architectures.įrom a reinforcement learning perspective, StarCraft II also offers an unparalleled opportunity to explore many challenging new frontiers. However, when trained on the main game, these agents are unable to make significant progress. On the mini-games, these agents learn to achieve a level of play that is comparable to a novice player. We give initial baseline results for neural networks trained from this data to predict game outcomes and player actions.įinally, we present initial baseline results for canonical deep reinforcement learning agents applied to the StarCraft II domain. In addition to the main game maps, we provide a suite of mini-games focusing on different elements of StarCraft II gameplay.įor the main game maps, we also provide an accompanying dataset of game replay data from human expert players. We describe the observation, action, and reward specification for the StarCraft II domain and provide an open source Python-based interface for communicating with the game engine. It is a multi-agent problem with multiple players interacting there is imperfect information due to a partially observed map it has a large action space involving 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 requiring long-term strategies over thousands of steps. This domain poses a new grand challenge for reinforcement learning, representing a more difficult class of problems than considered in most prior work. ![]() StarCraft 2 esports, which is part of the highest echelon of professional competitive gaming, will also continue going strong as it has been through our partners ESL Gaming and GSL.This paper introduces SC2LE (StarCraft II Learning Environment), a reinforcement learning environment based on the game StarCraft II. "We're not planning a Q4 balance update given that we did one a few months ago," Blizzard's Rob Bridenbecker wrote, "but as always, we do plan to continue doing them as needed in the future. Regardless, today's news marks the end of an era for one of Blizzard's biggest and most-played games, and another big change for the developer itself. Will Blizzard launch another RTS sequel? Or does the future of StarCraft like in another form or genre? We'll have to wait and see. In a blog post, Blizzard said the "good news" here was that it "will free us up to think about what's next, not just with regard to StarCraft 2, but for the StarCraft universe as a whole". ![]() StarCraft 2 will no longer receive new content, developer Blizzard has announced, just over 10 years since the game's original launch.īut the iconic real-time strategy game will continue to receive ongoing balance fixes and season rolls, and Blizzard will continue to support the game's professional esports scene.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |