It has come to my attention that a lot of people in the community deem it impossible for an AI to make a decent wall off at a ramp. Because making a wall off has always been on my TODO list from day 1, and I could not think of a good reason why it should be impossible, I have decided to accept this challenge.
Step one of this challenge is creating an algorithm that can analyze a ramp. I am not going to pre-program build positions based on the map name or ramp type. Because if I build something, I do it right and make sure it will work on every map. The AI should know how much space needs to be covered to make a complete wall off.
Step two would be an algorithm that calculates building locations based on building size and current wall off status of the ramp, so it can use any building that the build might require to continue the wall.
Step three is the race specific door mechanism. A Terran AI should have at least one supply depot in the wall that it can lower to allow passage for units. I will probably make the depot permanently lowered until enemies are approaching. The Protoss AI needs to leave one tile free as a doorway and place a unit on that tile on hold position (preferably a Zealot).
I have already worked a few hours on this challenge and I have completed step one. Here are a few screenshots demonstrating the algorithm. As promised, I have not manually placed the locators to match the ramp type in any way. This is all done dynamically.
Small ramp on Metalopolis
Big ramp on Metalopolis
My comment system was being flooded with spam comments. I have instated the rule that all comments need to be approved by me before they are posted on the site. However, once a user has 1 approved post, future posts of that person no longer need approval. So keep using the same username and have a little patience when it’s your first time posting. That should keep the site clean.
Sorry for lack of updates but there isn’t much interesting to say. I have been fixing minor bugs that are causing some error messages and working on performance issues here and there. For example, the individual unit control algorithm would get pretty heavy if the AI controlled 50+ zerglings and also had a 150+ supply army. Although none of the implemented unit micro tactics can do anything for zerglings, the fact they are checked anyway could cause some slowdowns. I have added an early out to prevent the most of these slowdowns.
I will release v1.00.02 soon because all those little bug fixes can still add up.
Today I added the ability for Zerg Overlords to fly away when they are under attack. The fact they were completely ignoring all incoming attacks looked kinda silly. I have also fixed a bug where the AI would think an army of Corruptors could take on a small ground force. Now it will recognize the fact the Corruptors can only shoot air targets so will now retreat.
I am planning to add unit kiting soon. Stalker play should become a lot more interesting. Not to mention Reapers. 
The maps have been uploaded on the NA Battle.NET servers. Enjoy!
I have uploaded new EagleAI maps to Battle.NET. They contain all the fixes I made this week. For more details, check the Change Log page.
I’m still having problems with units getting abandoned and forgotten on the field, because I didn’t have much time for extensive testing this week. I will be depending on the bug reports a lot more since I no longer can spend half a day watching games and writing down stuff I need to fix. So please play with the new version and share your experiences, the good, the bad and the bugs.
I have send the maps to a US player who is also a regular on this site. He should be able to upload the maps so the US players can enjoy the AI on their Battle.NET as well. I will let you know when it’s done.
Got another few hours in today. First thing I did was fix the “unable to find main building location” error spam on Xel’Naga Caverns. It was caused because the algorithm tries to find a place for a Command Center when the rocks get destroyed to determine the main building location. Since Command Centers cannot be placed on creep, I checked the location for creep first, and used a Hatchery for the check instead if creep was present. However, if the exact spot that was tested did not have creep, but creep would still interfere with a Command Center placement, the errors would appear. Yesterday I realized that Hatcheries can also be placed on non-creep terrain, so I simply replaced all Command Center checks with Hatchery checks. Now it doesn’t matter if there is (partially) creep present during the check.
I have removed some limitations on the expand code to try to get the AI to expand more when left alone. This should solve the problem where turtling can still be effective because the AI won’t take advantage of the fact you are not leaving your base.
I also modified the attack approaching code to take into account a walled-off base. It will now be able to find your base, walled-off or not.
I am keeping a more detailed change log from now on, so you can check all the changes (major or minor) for each release on this site.
I have started working on the bugs from the TODO list today. I have fixed the stupid worker transfers that people have reported. The AI should now also use workers to kill minor threats in the mineral line. A single zergling/marine/zealot will no longer be able to get x worker kills without being noticed.
I also worked on better anti-air defense and I fixed some other minor bugs I encountered while running today’s test matches. Again I have noticed multiple units sitting around on the map without being assigned to any group. I think this behavior is the cause for many late-game bugs like unwillingness to attack. I’m trying to figure out how those units get “lost” so I can fix it.
About five minutes ago, I put the final touches on the big graduation hand-in for tomorrow. This means I will have time to work on the AI again!
Not today though. I think I deserve a little R&R after the week of hell. I will however upload the script files of the current AI as a download so the people outside of the EU region can continue playing with the AI through the launcher. I am currently working with Refnil to get the EagleAI maps hosted in the US as well.
I have gathered all the feedback people have been giving me about the AI currently online and I am very grateful. A lot of people are challenged by the AI and are having a good time playing against it. But there are still some bugs left (especially high-level players encounter those) that I need to fix soon. I will probably update the maps on Friday with fixes.
Thank you for your support and please keep in mind that although no longer obliged by school to work on this project, I will continue to do so because I sincerely enjoy it.
Everyone, I need your help! This AI project is my graduation project for school. I am currently writing the final report and I need to know the communities opinion of the AI. When I started on the project one of the goals was to make it challenging for the majority of the playerbase. I need to know if I achieved this goal.
Please play against the AI on the maps I uploaded (preferably more then 1 game, the AI knows a lot of different strategies) and fill in the form below:
1. What 1v1 league are you currently occupying? (if you don’t agree with your current league, tell me which league you should be in)
2. How many games did you play against the AI and on what difficulty? (remember the AI does NOT cheat on Insane)
3. Were you challenged by the AI?
4. Do you consider this AI to be an improvement over the AI that comes with the game?
5. If you answered no to one of the two questions above, would you say this AI has the potential to become better?
6. Would you consider the AI to play more like a human instead of a computer? (especially noticeable on lower difficulties)
7. Did you enjoy the match(es)?
I need the answers by tomorrow, so please help me out if you can! You can post the results as a comment or mail them to eagle@eagleai.net.
Thanks!