I think we are getting pretty close to a cannon rush bot. A cannon rush bot could easily beat the built in AI.
We need to add a gitignore file to make it easier to avoid checking in binaries or that workspace.xml file or something.
What is the shortest sequence of chokepoints to travel through to reach a specific bwta region? What is the shortest path between two points in the same bwta region?
Can this unit travel to this point? If I place this building type here, will this unit still be able to travel to this point? What is the best position for these three buildings to minimize the gap distance on this chokepoint?
Find ways to increase worker mining speed. Also find a formula to predict a hypothetical mining rate for a given set of mineral positions, number of workers, etc.
It's probably too much to ask to set up a starcraft server to do runtime tests on our code, but we could do a simple compile test to confirm we don't break the build. I'm not certain how nicely something like travis-ci plays with intellij projects.
At some point, it will become convenient to have the camera automatically track the most interesting thing on the map so its less work to watch the bot play.
Many AI's get easily distracted when there is a worker running around the mineral line and you can bring their economy to a standstill.
We will have to defend against harrassment against our mineral line. This should also include ling rushes. A 4 pool rushbot is like the Hello World of bwapi bots. So adding this would immediately eliminate the threat of a significant percentage of bots in the tournament.
BWTA2 has some irritating problems. I'd rather use BWEM. dgant is working on writing a JNI wrapper. We could help with that.
The build manager should use the build planner to speculatively plan out the positions of all of the buildings in the build queue.
We will also need a decision-making layer to generate items in the build queue. UAlberta uses this thing called BOSS to search for an optimal build order. We could either figure out a way to use it or write something that does roughly the same thing.
This could use potential fields (ala Ironbot, or Berkeley Overmind) or combat simulation (ala UAlbertaBot). Combat simulation seems like the most accurate approach, but it seems complicated and computationally expensive.
builders get stuck or confused all the time. The builder should detect why a builder hasn't started building and try to fix it. Perhaps eventually give up and reassign back to the mining manager and enqueue your building type back into the build queue.
- Gitter chat room is up!
- Basic base building planning is now possible
build planner is usable but needs to add support for geysers and additional planning directives. Also spiral search doesn't make sense for all directives.