search

google-deepmind / lab

A customisable 3D platform for agent-based AI research
Other
7.11k stars 1.37k forks source link

How to create procedurally generated levels #67

Closed hirods21 closed 6 years ago

hirods21 commented 7 years ago

I'm doing some experiments to reproduce the results in "Reinforcement Learning with Unsupervised Auxiliary Tasks." https://arxiv.org/abs/1611.05397

Does anyone know how to create procedurally generated levels (e.g. nav_maze_all_random 0{1, 2, 3} in the paper)? In the settings, a new maze (new geometry, textures, goal, and start) should be generated on-the-fly at the start of each episode.

ghost commented 7 years ago

Hey! I'm facing problems trying to do procedurally generated levels. I just posted an issue here https://github.com/deepmind/lab/issues/73 .

Have you been successful in your endeavor? If so, may you share some insight with me?

Thanks.

hirods21 commented 7 years ago

Hi, I haven't solved this problem, but the solution in your post looks nice. I'll try something around it, and let you know if solved.

tkoeppe commented 6 years ago

The nav_maze levels were generated with different tools. Note that the Quake3 .map format is standard and many tools exist to create maps, so you can take your pick. The built-in "text level" API is fairly simple and can only generate rectangular mazes with "thick" walls.