Digital game AI Technology(spatial recognition technology)
This section summarizes the basic technologies of digital game AI discussed in the previous sections. The reference book is from “Textbook of Digital Games: The Latest Trends in the Game Industry You Need to Know“.
The quality of a digital game character’s AI depends on how much control it has over time and space. This also means how well the AI can recognize the surrounding environment and how well it can construct actions within a time range and time scale. I will describe the corresponding technologies for each of these.
First, let’s talk about space and object recognition. The field of digital games is much more complicated than that of Go or Shogi in terms of flight. In general, to represent such a space, we use a method of laying out points that are indicators of location (WayPoint) or triangles (Navigation Mesh), and connect these elements to form a network graph. These elements are then connected and processed as a network graph.
The most common processing of network graphs is path search to find the shortest path, and algorithms include Dijkstra and A* search.
In general, when an AI that exists in a certain world recognizes that world, it needs to be given a form of knowledge about “things”, “world”, and “events” so that it can interpret that world. It is very difficult for an AI to know what to extract from the world and how to interpret it on its own, and in the case of game AI, the developer gives it to the AI.
The most important knowledge representation for digital game AI is the World Representation, which is the counterpart of the entire map. The network graph described above is the world representation of the terrain for the AI, and in order to make the AI more intelligent, it is necessary to enrich this knowledge representation and world representation rather than the thinking algorithm. For example, by adding additional information to the network graph, such as the brightness of a point or how far away it is, we can enrich the AI’s perception itself and make it capable of developing a variety of thoughts.
In game AI, there is no fixed form of knowledge representation, and the design and development process begins with searching for the appropriate form of knowledge representation for each game. For example, in the case of enriching spatial information, just laying it out as a single layer only allows for detailed thinking in terms of locations, but by grouping locations in clusters of a certain size, it is possible to express a single coherent domain. (Technology to hierarchize space by scale)
The information about possible actions in a space (object) is called affordance. By filling the space with the affordance information, Ai can create a list of possible actions when it enters the map, and choose the best action from it.
コメント