Casting Shadows

BF Wolfe, an old friend who is the sort of cool person other people have to roleplay to even approximate, received some inspiration from our recent reminiscences about Battletech, and set out to investigate to determine – all external factors being equal – which traits held the most consistent value in mech design.

I’ve always wondered how balanced the original Battletech was, and I’ve decided to bring one of my favourite AI algorithms to bear on the problem. Anything you can spot in this reasoning to improve the process, let me know. 🙂

BF Wolfe

Here is a brief description of the ‘genetic algorithm’ BF ran first:

• Begin with a pool of 50 completely random mechs in the first generation.
• Create a ‘fitness function’ which evaluates these mechs against each other in a round robin battle.
• Select the top 20 ‘mechs to move on to the next generation.
• Add 30 random variations of these mechs. Mutations (minor random tweaks) and crossover (random hybrid) are the two most common variations.
• Start the next generation with the new ‘mechs and repeat until you have some stability

Assumptions & Conditions:

• Ease and speed of programming must be maintained throughout because we work for a living
• Only minor mutations for lack of an easy way to program hybrids of different weight ‘mechs
• Each mech fires every weapon in range
• Heat from this weapons fire is calculated, but not avoided
• Distance between ‘mechs is determined randomly each firing cycle
• Speed is a full run for each firing cycle
• Battle ends with the loss of the Centre Torso, or Head, or from a Heat Shut Down
• Initial results will be used to guide additions to further realism in the battle conditions

The Investigators:

Once results were obtained, the next phase would be to determine the value of the result, and the influence of the simplified assumptions made in the fitness function. Once the results of the first trial were in, Douglas, another old friend and Battletech aficionado, was called in to test assumptions and broaden thinking.

Douglas on Douglas:

My favorite mech was the Wolverine at 55 tons. Later versions of Battletech had a similar mech filling a similar role, and there was a 50 or 55 ton model with two PPCs that became my new favorite in the last Mechwarrior computer game I played.

I generally felt that in a free-for-all, use of terrain, mobility, armour and speed (but not sacrificing armour for speed) worked well for me. The two 50 to 55 tonners I mentioned had just enough armour to take a couple bad hits, speed, jump jets, and good heat management. Weapons that didn’t run out of ammo quickly (or at all) and having good heat management meant more in a prolonged fight that one or two turns of glory.

Me on BF’s Tactics:

BF’s approach was much like Douglas’,  motion, clever use of terrain, and cheerful use of enemy piloting errors. It was like being slowly kicked to death by that big, yellow, “Have a nice day” smiley face from the ’60s. Few people can make you enjoy getting the crap blown off your ‘mech like BF.

He seemed to favor relics and lost causes, and took to the field in ‘mechs that I considered to be walking ammo explosions, or mobile gunnery targets, but always seemed to comport himself well.

Douglas on my Tactics:

While I mixed it up with hit-and-run tactics, evasive maneuvers, and sudden charges—often jumping all over the place being very hard to hit, Runeslinger had different tactics. He used intimidation tactics with a Warhammer, a walking artillery-piece with shorter-range weapons just in case. He would often sit on a hill and rain death from afar. Other players would try to go after me, probably seeing the tonnage difference as making me a weak, easier target. If a single player came at him, he’d change his tactics and the Warhammer was versatile enough to do that.

Usually in a free-for-all, I’d get banged up taking out my first two or more heavy or assault mechs, and then players like Runeslinger would rain death down on my location before moving in on each other. Enough concentrated fire at long range would finish me off.

Initial Results:

Once results were obtained, the next phase would be to determine the value of the result, and the influence of the simplified assumptions made in the fitness function. The first shot ran for500 generations, and ended up with each top ‘mech being 90 tons, and armed with 1 ac10, and 7 medium lasers.

Trial 1: The Top 20’s Best, Median, and Worst ‘Mechs

Best

Mech 0 From generation 480 with a maximum generation streak of 20

weight = 90

engine = 270

walk = 3

run = 5

jump = 0

Armour = 21.5

Weapons :9

AC10, Medium Laser, Medium Laser, Medium Laser, Medium Laser, Medium Laser, Medium Laser, Medium Laser, Medium Laser,

weight total =  cockpit: 3.0 internals: 9.0 engine: 14.5 gyro: 3.0 weapons: 19.0 Armour21.5 Heat Sinks: 20.0 Jump Jets: 0.0 total: 90.0

Middle of the Pack

Mech 10 From generation 495 with a maximum generation streak of 5

weight = 90

engine = 270

walk = 3

run = 5

jump = 0

Armour = 21.0

Weapons :10

AC10, Medium Laser, Medium Laser, Medium Laser, Medium Laser, Medium Laser, Medium Laser, Small Laser, Medium Laser, Medium Laser,

weight total =  cockpit: 3.0 internals: 9.0 engine: 14.5 gyro: 3.0 weapons: 19.5 Armour21.0 Heat Sinks: 20.0 Jump Jets: 0.0 total: 90.0

The Cut-Off Point

Mech 19 From generation 499 with a maximum generation streak of 1

weight = 90

engine = 270

walk = 3

run = 5

jump = 0

Armour = 21.0

Weapons :10

AC10, Medium Laser, Medium Laser, Medium Laser, Medium Laser, Medium Laser, Medium Laser, Medium Laser, Medium Laser, Small Laser,

weight total =  cockpit: 3.0 internals: 9.0 engine: 14.5 gyro: 3.0 weapons: 19.5 Armour21.0 Heat Sinks: 20.0 Jump Jets: 0.0 total: 90.0

Now what?

Looking at the results, we naturally fell to sharing old stories, describing each others’ tactics, and trying to remember past glories in favorite ‘mechs. It was hard at first – at least for me – to know what to do with the results once they had appeared.

I offered the following as my first impressions:   (Favorite Battlemech: Warhammer)

In my experience with Battletech, a player cannot go wrong with a host of medium lasers so this feels right. I find that if variants are allowed, one of the first things people I know learned to consistently get rid of were small lasers, so its disappearance in the evolutionary cycle is something I would expect.

A heavy weapon which does 10 points of damage reliably at medium and long range is an efficient choice. The heat versus ammo trade off is something which can be debated endlessly, so seeing the balance between a very heavy punch at medium range, and a decent, low-heat punch at longer ranges seems like a good result.

Perhaps the biggest surprise for me was the tonnage.

Each ‘mech has strengths and weaknesses as built, and each player has preferences which will either augment or exacerbate those traits, so the idea of finding a way to refine a sense of exactly what makes a ‘mech great, when taking the pilot out of the consideration is very useful.

As our discussion has shown, I am very much in favour of having a lot of medium lasers, and approve of that evolution in the sequence you observed. Doug, however, cautions that for larger mechs, a heavier weapon will be necessary, and further cautions that the suite of medium lasers will prove better in theory than practice. I totally agree, and my ‘mech choices and variations reflect that agreement, not the theory… However, I still very much believe the theory, if only because some people have good luck,  while others don’t.

As it happens, your fitness function provided exactly that result: a suite of mediums and 1 heavy weapon. Interesting.

Douglas offered these thoughts:  (Favorite Battlemech: Wolverine)

Regarding weapons, one or two medium lasers as backup for extra short-to-medium range firepower was a good idea. Larger mechs based solely on nothing but medium lasers (ostensibly the highest DPS output) seemed to not work as well in practice as in theory. You’d miss the lack of a more versatile large laser, PPC, or the like. Medium lasers made decent primary weapons on light mechs.

I placed equal importance on weapons and armour—perhaps more on weapons since killing your opponent faster amounts to the same thing as more armour. However, heat management was always an issue. Extra weapons that you could fire once, overheat, and then not be able to use for a while didn’t work out well.

In terms of raw power, nothing beats going for higher tonnage, but I didn’t like that. I enjoyed taking out heavy and assault mechs with my 50 to 55 tonners, just for the challenge.

We tended to notice that the PPC, despite its heat, was an awesome weapon. It had a nice punch, great range, and didn’t need ammo.

Any weapon that did a lot of damage to a single hit location had a greater chance of getting crits sooner or taking off limbs altogether. Scattershot weapons were better than nothing, but seemed to take longer to get through armour (sprinkling damage all over the place instead of punching through). Although, the AC 20 was desired for this damage potential, it’s limited range and limited ammo made it a poor choice for all but the biggest of mechs and even then you needed a slow target at close range.

I guess the running theme here is that raw Damage-per-Second comparisons are less important—in my opinion—than comparing the ability to successfully apply that damage. If you are out of ammo, overheating, out of range most of the time, or just not able to hit those “normal” 50-ton mechs that keep moving around too much and making surgical strikes, then the DPS will mean a lot less.

BF kept things moving with these comments: (Favorite Battlemech: Dervish)

I’ve been doing multiple runs while varying different conditions, and the things that show up as important are very telling. For example, gunnery skill:

In a competition with Medium ‘mechs only, when I run the test with Gunnery Skills of 4, I get ‘mechs with 1 large laser, a few medium lasers, and a run of 8.

• One run with these stats practically duplicated the Phoenix Hawk.

Lower the gunnery to 2 or 0, and the optimal ‘mech has a run of 3! with LRM5’s, a PPC, and 5+ medium lasers, with the heat sinks to fire them.

Testus Interruptus:

It seemed like we were getting closer to real results. As BF controlled the very forces of the universe in his virtual pocket realm, Douglas and I waited with baited breath for the results of the heavy and assault trials. Would they turn in different variations than we had seen in the Medium class? If they did, what would that mean?

Questions our discussions raised:

• What effect will altering piloting and gunnery scores have?
• Why no jump jets?
• Is slower and heavier really better?
• Who would win in a bar-fight with broken glass and lava on the ground?

With the initial results already causing good discussion on key factors like damage capacity versus appropriately-ranged weapons per ‘mech classification, I pretty much feel like the experiment has already produced interesting and beneficial results, but more awaits!

Stay tuned to find out what happened next~