JMoe: The Mastermind

I was hoping to have more information for this post, but I’ll have to settle for posting our team document. The Battlecode organizers never got around to updating the website, and so I don’t have the actual final results. You can grab the PDF at the bottom of this post. Overall our team worked out reasonably well. We needed more time to tweak and optimize behaviors, and a lot of the strategy just didn’t work out as well as it should have. Still, I’m happy to settle for winning an award and getting the credits for the class.

Battlecode Strategy Writeup

This is again a rather quick note. The developers (and sponsors) of Battlecode award a variety of prizes to teams for many different things. One was a prize for the most merciless (ie the team that hunted down and killed a defenseless, move and turn only) archon the fastest, while another was for the best team name. Blooregard Q Kazoo didn’t win either of those. We did however win the best formation award. This award was for our arrow formation which we would form above our archons to defend against EMP and air attacks. We also had the ability to fire out the center of the arrow, rather like a crossbow, or bow and arrow. We disabled this for the tournament, but it was coded and working. This earned us the best formation award, which was $200 and an iPod Shuffle.

The next post about Battlecode will likely be the last, and will contain our strategy document and descriptions of our matches (it may need to be two posts to avoid being too long). I will also hopefully have our final ranking from the tournament, so we can see if we improved (somewhat unlikely), stayed the same, or fell from our ranking in the seeding tournament. I feel that it is likely that we lost some ground, but probably not a lot. I know we didn’t finish in the top 16, which doesn’t leave a large margin to have improved, but it is probably possible.

I feel compelled to at least write a quick post noting that I am not in fact dead, but have been busy with Battlecode (see a couple posts down) for the past several days. Although the final seedings aren’t out yet, my team did reasonably well in the qualifying tournament, losing to two quality teams. We managed to win our first match against team General RAAM (after a bye), lose our second match against team Battletoads (the top ranked scrimmage team, seeded 9th), win our third match against team Global Warming, and lose our fourth match against team Little (to the third ranked scrimmage team, seeded 8th, beat us in seeding tournament). Battletoads was also one of the teams that finished in the top 8, so I don’t feel too bad about losing to them at all. After the final tournament (on Saturday) I will post our teams strategy writeup, and also hopefully go through our four matches, with some commentary on why we lost, and why the other team deserved to win.

With the basics out of the way, it’s time to delve a little bit deeper into some strategies for Battlecode. These are a few that have all been used by multiple teams, so I’m not giving away anything top secret here (especially after the seeding tournament).

• The Scout Tanker: With a low upkeep cost, scouts can be used to provide Energon to other units with a higher upkeep (EMP’s especially) to keep them alive longer. This can help to greatly extend the range of EMP’s. One of the most interesting applications of this could be called the Scout Railroad. It was a line of 10 scouts with one end anchored over the archons to pass up energon, who would feed any EMP that traveled along the rail to keep it alive. It wasn’t terribly effective, but looked impressive.
• The Archon Trap: Archons are a very difficult unit to destroy. Their health regeneration is able to match the damage output of a soldier with no problem, and can even survive two attackers with no difficulty if it is moving. The trick is to trap an Archon so that at least two, preferrably 3 soldiers are able to attack it. This can be done in corners, or with a large group of scouts along a wall.
• The EMP Wave: EMP’s are a dangerous offensive unit, and some teams simply used them as their only major unit. They would use Scout Tankers to keep the EMP’s alive, and send wave after wave of them at the enemy, hoping to cripple their energon production, and win that way.
• The Tank Assault: Tanks are a powerful ground unit, dealing major damage, plus splash damage. A pair of them can take down even Archons with little difficulty if they have a spotter. Teams that can move their archons, and locate the enemy without taking too much damage, and can get in range of a tank strike can be deadly.

There are of course many other tactics and strategies, but I don’t want to give away everything (as I do have a few ideas that I didn’t see implemented at the seeding tournament). Our team did quite well, being seeded 24th out of 137 teams. Hopefully that number can be improved in the final tournament.

Battlecode (also known as 6.370, or in past years, Robocraft), is an annual AI programming competition that occurs every year during IAP (Independent Activities Period). It centers around writing a program that will control a variety of different robots to achieve a certain goal. In previous years it has ranged from kill the queen, to capture the flag, to king of the hill. There are a variety of different robots, which change every year. The robots this year are:

• Archons: The most important units, the only unit with positive energon production
• Soldiers: The basic ground unit. Attacks other ground units adjacent
• SAMs: The anti-air ground unit. Attacks only air units with a range
• Tanks: The ranged attack ground unit.
• Scouts: The basic air unit. Attacks other ground units adjacent
• EMPs: An evolved air unit. Has a suicide ability that reduces the energon production of Archons in a radius. Can also attack air units.

Each of the units that isn’t an Archon has an upkeep cost in energon. Energon is the energy of the game, and it allows your units to live. Attacks reduce energon, and EMPs reduce the energon production of Archons. The goal of this year is to kill the 4 enemy archons, or to satisfy one of the other victory conditions. The victory conditions are:

1. Largest number of surviving Archons
2. Largest total energon production among Archons
3. Largest total energon among surviving units
4. Lowest team number (so earliest registered)

There are, of course, a few complications to make writing the AI more complicated than simply that of programming a basic game AI. There is no overaching player. In Battlecode, each robot runs a copy of your code, and must work together with the limited information available to each unit. This includes their sensor information (which varies based on robot type). They also can message robots in a limited radius to share information. There is no overarching “super” player that can see the entire map. You have to determine what the map looks like, how to pathfind, and everything on the fly with limited information.

With the basics of Battlecode out of the way, you can expect another post tomorrow explaining a few ideas for strategies (some I might use, some I probably won’t), and any details that I realized that I missed. The full game is of course more complicated, but this should be a sufficient overview for what I explain tomorrow.