Reinforcement Learning PCG

摘要: All the way back in Chap. 2 we discussed search-based PCG (SBPCG). This was an approach to generate game content by authoring a space of possible content, away to move through that space (metaphorically, literally it involves editing some current piece of content), and then some way of evaluating the content (a quantitative measure of content quality or whether it satisfied some set of properties). This allowed a generator to start from some initial (empty, random, etc.) piece of content and make changes to reach some final output content. As with all classical PCG approaches, this approach allows for a great deal of control, but also requires specialized knowledge to apply effectively and a fair amount of authoring effort. Unlike other classical PCG approaches, it also can be very time-consuming due to the iterative changes of the search process. Back in Chap. 4 we gestured at the existence of PCGML …