This is an interesting idea.
Many turn-based strategy games involve a lot of luck, so the warbot could also be programmed to take the enemy's turn by means of manipulation. For example, moving one of your units might make the AI do something totally different if done one frame later, causing the entire branch to be "incorrect".
I think you're right about that.
But, in two of the four games I mentioned (Tactics Ogre and Shining Force) the RNG is impossible to manipulate by conventional means, such as waiting a few frames, or holding other buttons while pressing A. I assume that would keep the possibilities relatively low.
Front Mission on the other hand, is almost as manipulable as Pokemon, IIRC. Though, adding luck wouldn't make a branch "incorrect"; it would make a separate branch.
...Or rather, infinite branches, one for each possible frame of waiting. But, I think that would be solved by having the AI read the RNG's memory address and picking the optimal frames on which to attack.
well, for every frame, there's 2^8 input possibilities, or 256. If you take a video that's 10 minutes long, that's 60fps * 600 sec = 36000 frames, giving you a total number of 9,216,000 different movie possibilities. If you can cull the amount of control input, say, no up+down; or if you give it instructions to watch for certain RAM values- no death, don't pick up a certain item; you could reduce that number to something more manageable.
That's why I'm talking about turn-based strategy. It would only be necessary to consider a handful of decisions.
If a battle in Shining Force has one character and one enemy (which, granted, is unlikely) then you'd only need to consider... what? Two decisions?
And of course, the purpose of the bot wouldn't be to randomly peck at the controller; the inputs would follow some sort of logic, such as reading the cpu and ram to know when and what to input.
But, that's a bit technically advanced for me to even speculate about at this point.
Warp wrote:
There's actually a factorial amount of them (with respect to the number of pressable keys), which is far fewer than infinite. ;)
Har har. :D
Ranting onward a little bit, I was thinking that a bot like this would probably work using a decision tree, in addition to random trial and error. The start of the battle is the "root" and decisions made for any member of the player's team is represented as a branch in the tree.
So, at the start of combat, the bot gets control of the leader, who can move to X possible locations, attack, pass his turn, or use an item. Each subsequent action adds to each branch in like fashion.
When some outcome is reached, the AI notes which branch was used, and what the outcome was and the time it took. Then it uses that information to weight the branch it just tried, and its sub-branches. It restarts and tries again, this time taking into account the weighting of each branch; it would make little sense to try the same course of action over and over if it always fails.
Hopefully by doing that, it could narrow down the field enough to make a battle with millions of possible combinations of decisions, into something a PC could trawl through in a few hours.
It's an idea I've been tinkering with.
Consider that there are essentially infinite combinations of keypresses in a game like SMB, per frame, and that it's impossible to brute force with AI. (Note that we're probably dealing with computational intelligence when we're talking about a TAS-bot.)
But what about a turn-based strategy game like... Shining Force, Tactics Ogre, or Front Mission? The decisions are far fewer. Move/attack/use item.
So, if the bot is designed to use a few basic strategies, it might be able to simply try a whole lot of possibilities, most of which (hopefully) are not hopelessly insane.
I'm still just tinkering with the idea, but I'd like to look into this more deeply after I've purchased a new computer, a few months from now. I'm planning on getting something fairly high-end, to replace my current 500mhz PIII, which can't really do much anymore. But before I get too carried away with this idea, I'd like to know a little more about what I can expect out of a new computer.
What sorts of framerates do you guys get? I can barely get 60fps in most games, using Snes9x, if even that. If a modern machine is, say, 15x faster and also assuming that the AI is computationally inexpensive... that's a lot of trial and error.
Any thoughts on this wacky, impractical idea?
Somehow, this is less interesting to me than the other two DD games. It's a good movie, though... I like how Hanzou is talking while being maimed by claws.
Yes vote.
It took about fifteen years, but I've defeated Smash TV.
I have to say, it's a perfectly mediocre movie. I didn't annihilate the game effortlessly, and I didn't scrape by perpetually on my last life. I'm pretty impressed with myself, though. Cobra death AND the end guy without dying... I think.
(And that thing I did at the end doesn't count.)
Anon wrote:
Honestly, the most ideal Panel game for TASing sounds like Crazy Hard mode of Pokémon Puzzle League's VS/Challenge mode. From what my friend tells me, it's more evil as the Intense difficulty from Pokémon Puzzle League (constant ×8 and ×9 chains), only without the benefit of the computer having a health bar.
Interesting. Maybe I'll give that a whirl if I ever get my hands on a computer that can actually emulate anything beyond SNES. :)
Perhaps some easter eggs for detected sequence breaks, or even stuff "left in" the game with no known means of access at ship time...)
That's awesome. Every developer ought to know by now that the fun of gaming is not jumping through predefined hoops, but slipping through those prison-cell bars. :)
Given this, what engine mechanics and gameplay features would make the game most worthwhile to TAS?
Personally, I like choreographing fights. Give the player situations that would be nearly impossible to do in realtime, like recreating the "Neo vs 100 agents" fight from Matrix Reloaded.
The key would be giving the player a broad range of attacks, sufficiently raising the complexity to the point where TAS-tools are able to exploit the engine where player's lack of reflexes and wit tend to fail.
Watching parts of this movie, even parts as trivial as Samus bouncing backward across the big floor of spikes (to ride the statue- you know, that part) it dawned on me how meticulously everything must have been planned.
My only complaint is that there aren't more opportunities to do that speedboost-ram thing to kill bosses.
Smash.
Wait... so the forces of evil held off for a couple years so that the hero could have a kid with some chick who joined him in scaling a couple towers...
The forces of evil were smoking pot and playing mortal kombat in their mother's basement. That's why they got thrashed without the heroes making any attempt at power-leveling.
And Selan isn't 'some chick'. She's a 'magical wife'.
PDP TASed would just be "input B-A-L-L at start, do small (7x, usually less) combos to fill up enemy playfield, win in five seconds" for every round (Acmlm was doing a TAS like this of the US version.)
:/
Yeah, it occurred to me after I posted the suggestion. I think I might take a stab at the final round- nearly lose at first while setting up a massive combo, and then brutally savage the opponent.
-
Here's an (admittedly shabby) attempt at Tetris.
My lust for Tetrises was my undoing.
Panel de Pon, played on Hard difficulty. I do relatively decent until the final boss appears, and then it goes to hell from there. I'd say nearly half the movie comes from trying to beat her.
Wow. I haven't seen the boss yet, but this is pretty cool so far. Have you considered doing a TAS?
Unlike RCR, I can actually play NARC fairly well. Made it all the way to the end without making a complete ass of myself... until the initials, anyway.
Movie version says [o1], but I don't know if that means [!] will desync.
Ah. Cool. It looks pretty solid.
I've always felt this game was ideal for TASing. It's fast-paced; the battles are quick, well planned, and seldom repetitive as one would expect from a 3-hour RPG movie.
...And it's fun figuring out how the puzzles work as they're being solved. :)
Yes vote.
River City Ransom
Here's me getting my face smashed in for three hours. The ~30 rerecords were for the twin fight, which was unholy.
Very, very boring, but it makes me smile to think that my insomnia might end up curing someone else's.
Edit: there was a weird glitch. I think it happened around 260000.
FSD? Flying spaghetti demon. Duh. (Or, look at the home row on your keyboard.)
Was it my imagination, or was that 1UP#* thing with the goomba an allusion to the river Styx? Senility, then death, then... dancing?
Nevermind.
Capcom could teach Valve a thing or two about physics, as Haggar (clumsily) demonstrates upon Belger.
Final Fight Guy (presumably J) using Zsnes 1.50.
Bwahahaha.
