Continuous Procedural Network of Roads Generation using L-Systems and Reinforcement Learning

Procedural content generation methods are nowadays used in areas such as games, simulations or the movie industry to generate large amounts of data with lower development costs. Our work attempts to fill a gap in this area by focusing on methods capable of generating content representing network of roads, taking into account real-world patterns or user-defined input structures. At the low- level of our generative processes, we use L-systems and Reinforcement Learning based solutions that are employed to generate tiles of road structures in environments that are partitioned as 2D grids. As the evaluation section shows, these methods are suitable for runtime demanding applications since the computational cost is not significant.