brittcyr / battlecode-caddar Goto Github PK

View Code? Open in Web Editor NEW

0.0 3.0 0.0 38.27 MB

Battlecode 2014 Team

License: Other

Java 97.41% CSS 2.15% Python 0.45%

battlecode-caddar's Introduction

battlecode-caddar

Battlecode 2014 Team

Final submission is teams/team050

battlecode-caddar's People

Contributors

Watchers

battlecode-caddar's Issues

Inconsistent state

If a pastr/nt dies, clan doesn't know to rebuild it.

hq computation

make hq compute especially during early turns

Create table of bots

Once the scrim server is set up, we will want to know which bots to not include for certain. Also, it would be helpful to have a description of where each bot leaves off

Scrimmage Server

Create a continuously running scrimmage server to help evaluate our bots

investigate other nt strategies

We should not necessarily be pulling in as fast as we do

Intended behavior of enemy radio tracking

Consider this conundrum:

team050 v doNothing on siege

COMPARED TO

team050 v pastr_attacker on siege

against doNothing, bugnav works normally, but against pastr_attacker, they clump in a corner because they are attracted to the enemy's radio transmissions.

We need to decide what the expected behavior is for attacking on enemy radio transmisisons.

search for nav fix

We should return a map if the target is a coarse match, not just exact match

Distracted bug

Should terminate bug if it gets too far off course or loops

Determine if a robot is alive

This issue also includes reinforcing the defense units when they die and the clan logic to create raiders after the defender clan is full

Implement Fibonacci Heap

create a fibonacci heap or something similar to improve dijkstra

Investigate D* navigation

see if that could help. at least we should be doing some kind of memory of what has been seen to help other robots

Naive first implementation was at about 200 turns to compute 30x30, the optimized fibheap was finishing at turn 55, the current "stringHeap" finishes on round 31 with about 9 of those rounds being from sensing the map and 2 from initialization that I can avoid by compiling properly.

So it runs in ~20 turns now. I have not checked correctness in a while, but if it is, I still see some improvements possible. This if it works would be a candidate for general navigation of cowboys if it got a little faster.

Exception

 [java] [A:SOLDIER#3642@868] java.lang.ArrayIndexOutOfBoundsException: 4869
 [java] [A:SOLDIER#3642@868]    at team050.rpc.Clans.getClanMode(Clans.java:61)
 [java] [A:SOLDIER#3642@868]    at team050.CowboyRobot.doAction(CowboyRobot.java:244)
 [java] [A:SOLDIER#3642@868]    at team050.BaseRobot.loop(BaseRobot.java:30)
 [java] [A:SOLDIER#3642@868]    at team050.RobotPlayer.run(RobotPlayer.java:37)
 [java] [A:SOLDIER#3642@868]    at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
 [java] [A:SOLDIER#3642@868]    at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:57)
 [java] [A:SOLDIER#3642@868]    at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
 [java] [A:SOLDIER#3642@868]    at java.lang.reflect.Method.invoke(Method.java:622)
 [java] [A:SOLDIER#3642@868]    at battlecode.engine.RobotRunnable.run(RobotRunnable.java:65)
 [java] [A:SOLDIER#3642@868]    at battlecode.engine.scheduler.ScheduledRunnable.run(ScheduledRunnable.java:37)
 [java] [A:SOLDIER#3642@868]    at java.lang.Thread.run(Thread.java:701)

Debug Dijkstra

It seems to favor direction 7, the last direction seen for some reason

Cleanup branches

it wont let me push now

Implement defender

play defense

IRC scraper

We should be monitoring the IRC channel

Detect engagement

This could happen when we update the state of a robot. We need a cheap and amortized way of determining if we see any enemy cowboys and need to fight

Pastr mercy kill?

(12:01:23 AM) [PaddleGoats]u_: lol duck your pastr distress signal is a little too late
(12:01:52 AM) The_Duck_: it's not calling for reinforcements
(12:01:57 AM) The_Duck_: it's calling for a quick death
(12:02:31 AM) jiangty: "kill me plz"
(12:03:01 AM) The_Duck_: I have a function in my soldier class called "mercyKillPastrs()." But the pastrs have to ask for it first
(12:03:10 AM) jiangty: rofl
(12:03:20 AM) jiangty: wait I thought denying was nerfed
(12:03:31 AM) The_Duck_: you can still shoot your own pastr
(12:03:37 AM) jiangty: and they don't get the milk?
(12:03:42 AM) The_Duck_: right

Noise Towers

Investigate the usefulness of noise towers

Functions to send and receive paths

Dijkstra is almost ready. We need to post and receive the paths computed at pastrs

Final Report

Need to write this by Sunday night.

Integrate new dijkstra

Macro Strategy

We need to formulate a big strategy

Also, note that there is no longer an advantage to suicide your pastr

Introduce memory and messaging into robots

Make the robots able to start to learn from each others paths

Implement raider

we need to attack

Fix height and width on map editor

They do not work on the editor

Enable cowboys to self reassign clans

IDLE cowboys should be allowed to just leave and join a defender clan and injured cowboys should be able to retreat into an idle clan

Blocky

Cannot spawn and play on map Blocky. Throwin exceptions all over the place.

Improve bug and modularize

make bug and improved bug navigation into a method in a static class, so we can combine it into other bots

Better Pastr locations

Release last Clan

(n-1)th clan clan has capacity for 10 people and won't release newspawns from hq when in idle mode until they have 10, which will never happen.

expected behavior should that they release after like 3 or 4 newspawns. roughly.

Use sight

Short distance navigation should use sight

Enforce 1 HQ for each team on maps

Map editor does not enforce this yet

BFS

Write BFS for short range cowboy nav

Pathing Fail on Temple

Reproduce

team050 v doNothing on TEMPLE

Problem

On Temple, BUILDER clan tries go to build its stuff in corner. But when they hit the wall, 3/5 of the clan decide the fastest way to get two squares over is to bug around the entire rest of the map, splitting up the clan and leaving PASTR + NT with no defenders.

Expected Behavior

They should all run into the corner and defend the PASTR and NT.

Support for more more granular coarseness

We are locked into C=2 or C=3. It would be preferable for this to be a general field, as Dijkstra computations are still taking too long to compute.

Save bytecode

I would ideally like to be able to add new features without worrying about hitting the limit. This means that I am going to try to reduce by maybe 25% or so in bugnav and micro