Solving Super Metroid: the collaborative room-by-room strategy topic

2008-05-27 01:40:11 pm

Eschews avatars

I've been inspired by the speedrun team over on the Prime boards and I was thinking it was time we created a solid 'best play' route for the entirety of Super Metroid. Yes, this will make subsequent speedruns relatively boring, but I think this is really the only available step up in quality from Hotarubi's run. There are tiny speed tricks in nearly every room; now we simply need to compile our knowledge to create an optimized humanly possible route. I suspect the TASers will have a lot to say.

I was thinking we'd just start discussing strats starting in Ceres and moving on, using the standard route (which I believe is set in stone for now; Hotarubi's route but going through RB/Maridia instead of GB on the way to Tourian) and posting in the form of text descriptions or maybe smvs. Who's up for this?

Thread title:

072

2008-05-27 02:27:58 pm

coral to complement blue

I don't think anyone is even going for the new any% record besides me, and I haven't done anything on it in a while. The problem with beating :32 is definitely not the route; it just seems so difficult not to screw up at least a few seconds' worth on every segment.

hero of the day

2008-05-27 04:25:44 pm

I like this idea, but it has flaws. Pretty much any TASer could give you the absolute "best" room strategies for the route, but many of them would be illogical for a human runner to use. Hotarubi's run uses pretty much the best strategies for console play. Hotarubi's run also used some pretty risky tactics, so you would have to determine the player's skill level before deciding what strategy to use for any given room.

As for the route, I am not certain that it is the best one. How much time does the murder beam save on mother brain? The spazer costs a good bit of time to collect. Doing WS early is also a possibility if a solid phantoon fight using just missiles was possible. The red brinstar shaft climb can be done very quick using the 'hero shot'. The missile lake can be crossed with a continuous walljump or with bombs just like Hotarubi's low% run. Having the speed booster in WS does not save much time, and if you can eliminate the huge backtrack from norfair a good deal of time will be saved. Doing Kraid's area at the end of the run will save a chunk of time too. Basically I think that the TAS in-game route can be done by a human player faster than the route Hotarubi used, if played well enough.

I like this idea, but it has flaws. Pretty much any TASer could give you the absolute "best" room strategies for the route, but many of them would be illogical for a human runner to use. Hotarubi's run uses pretty much the best strategies for console play. Hotarubi's run also used some pretty risky tactics, so you would have to determine the player's skill level before deciding what strategy to use for any given room.

As for the route, I am not certain that it is the best one. How much time does the murder beam save on mother brain? The spazer costs a good bit of time to collect. Doing WS early is also a possibility if a solid phantoon fight using just missiles was possible. The red brinstar shaft climb can be done very quick using the 'hero shot'. The missile lake can be crossed with a continuous walljump or with bombs just like Hotarubi's low% run. Having the speed booster in WS does not save much time, and if you can eliminate the huge backtrack from norfair a good deal of time will be saved. Doing Kraid's area at the end of the run will save a chunk of time too. Basically I think that the TAS in-game route can be done by a human player faster than the route Hotarubi used, if played well enough.

ING-X

2008-05-27 07:42:08 pm

yeah

Possibly.

But since I suck at this game (and Metroid games in general) I definitely wouldn't try it. :P

Racer J

2008-05-28 10:33:20 am

Yea, i'm pretty sure hotarubi's run is optimized for console, although some small improvements could be made.

For human players like myself, you might want to check out my run, I don't do anything too superhuman (the only really difficult part is getting springball w/o grapple), but still managed a 1:08. With the same path through the game, it would probably be possible to get below 1:00, but I'm just not good enough. I might start practicing and do a new run soon. But really, a thread like this could go on for 200 pages, because everybody has their own strategy, and in this game two completely different paths through the game can produce almost the same game time, so there will be lots of arguing.

My 1:08...
http://video.google.com/videosearch?q=super%20metroid%201%3A08&ie=UTF-8&oe=utf-8&rls=org.mozilla:en-US:official&client=firefox-a&um=1&sa=N&tab=wv#

Yea, i'm pretty sure hotarubi's run is optimized for console, although some small improvements could be made.

For human players like myself, you might want to check out my run, I don't do anything too superhuman (the only really difficult part is getting springball w/o grapple), but still managed a 1:08.  With the same path through the game, it would probably be possible to get below 1:00, but I'm just not good enough.  I might start practicing and do a new run soon.  But really, a thread like this could go on for 200 pages, because everybody has their own strategy, and in this game two completely different paths through the game can produce almost the same game time, so there will be lots of arguing.

My 1:08...
[url=http://video.google.com/videosearch?q=super%20metroid%201%3A08&ie=UTF-8&oe=utf-8&rls=org.mozilla:en-US:official&client=firefox-a&um=1&sa=N&tab=wv#]http://video.google.com/videosearch?q=super%20metroid%201%3A08&ie=UTF-8&oe=utf-8&rls=org.mozilla:en-US:official&client=firefox-a&um=1&sa=N&tab=wv#[/url]

horscht(i)

Edit history:

horscht(i): 2008-05-28 02:33:02 pm

2008-05-28 02:27:15 pm

defrag in progress

One should start writing a code which uses algorithms for the Traveling-Salesman-Problem.
Beating SM (or any other metroid game) is much like this well known problem:
You have a couple of rooms you have to visit, e.g. to get necessary upgrades or fight bosses, in the shortest possible time.
Of course this set of rooms depends on the kind of speedrun you want to execute (100%, low%, any%).
Such codes are able to determine the shortest route quite well.

But there is one additional problem with Metroid: The Upgrades itself.
Normally the traveling time between rooms is fixed and the code tries to find the optimal route, but with all the upgrades in Metroid all this times change, due to new movement options and speed tricks.

A very systematic method would be as follows: One has to make a list of times he needs to pass a single room. This has to be repeated with different gear (with/without speed booster, space jump etc) and for all rooms in SM.
Some combinations can be skipped of course, because we already know that with some equipment is necessary to pass some rooms, e.g. supers to open green doors.
Now the code needs to check, which upgrades have already been collected and thus set the correct times for the upcoming rooms in order to really find the fastest route.
Did you get me? I don't know how to explain better...

After collecting all this data one will end up with a 3 dimensional tensor of numbers each representing the necessary time to pass a certain room with a certain set of upgrades.

Maybe an example: Wrecked Ship Entrance
Depending on the upgrades you have, there are several methods to cross the lake and reach the entrance.
(1) bomb jumps
(2) shine spark
(3) grabble beam

So you have 3 times for this room. Lets label this room "WS_1".

So we have:

ROOM	(1)	(2)	(3)
WS_1	30s	2s	20s

Shine spark is sure the fastest way, but does this make worth the effort of getting the speed boosters? Basically yes, because we all know that the boosters are useful in a lot more situations. But these are the kind of questions, which must be considered by the code.

€: Anyone here studying math or informatics? This work sure would fill a master thesis. ;)

One should start writing a code which uses algorithms for the Traveling-Salesman-Problem.
Beating SM (or any other metroid game) is much like this well known problem:
You have a couple of rooms you have to visit, e.g. to get necessary upgrades or fight bosses, in the shortest possible time.
Of course this set of rooms depends on the kind of speedrun you want to execute (100%, low%, any%).
Such codes are able to determine the shortest route quite well.

But there is one additional problem with Metroid: The Upgrades itself.
Normally the traveling time between rooms is fixed and the code tries to find the optimal route, but with all the upgrades in Metroid all this times change, due to new movement options and speed tricks.

A very systematic method would be as follows: One has to make a list of times he needs to pass a single room. This has to be repeated with different gear (with/without speed booster, space jump etc) and for all rooms in SM.
Some combinations can be skipped of course, because we already know that with some equipment is necessary to pass some rooms, e.g. supers to open green doors.
Now the code needs to check, which upgrades have already been collected and thus set the correct times for the upcoming rooms in order to really find the fastest route.
Did you get me? I don't know how to explain better...

After collecting all this data one will end up with a 3 dimensional tensor of numbers each representing the necessary time to pass a certain room with a certain set of upgrades.

Maybe an example: Wrecked Ship Entrance
Depending on the upgrades you have, there are several methods to cross the lake and reach the entrance.
(1) bomb jumps
(2) shine spark
(3) grabble beam

So you have 3 times for this room. Lets label this room "WS_1".

So we have: [table][tr][td]ROOM[/td][td](1)[/td][td](2)[/td][td](3)[/td][/tr]
[tr][td]WS_1[/td][td]30s[/td][td]2s[/td][td]20s[/td][/tr][/table]

Shine spark is sure the fastest way, but does this make worth the effort of getting the speed boosters? Basically yes, because we all know that the boosters are useful in a lot more situations. But these are the kind of questions, which must be considered by the code.

€: Anyone here studying math or informatics? This work sure would fill a master thesis. ;)

devonodev

2008-05-30 04:26:24 am

Devonodev: Other D

after reading your post several times, i finally understand :D
its a good method.

Sneaky Paradox

2008-05-30 07:11:54 am

I like the idea, and I can help as I've gotten 00:36 so I'm quite fast. The only thing I see is that there really aren't as many improvements that can be done in a single segment except for a few enemy or damage boosts that I've seen from Cpadolf's latest TAS.

Cardweaver

2008-05-30 07:43:49 pm

Oddly enough, I actually understood most of the algorithm talk.

I think that would definitely work, though you forgot CWJ in your example, which I'd imagine is somewhere between 2-4 seconds if the missile isn't there (not sure, never pulled the CWJ off).

horscht(i)

2008-06-09 09:18:41 am

defrag in progress

CWJ? Continous Wall Jump? I don't know all this abbreviations.

Actually, I was thinking about such a method half a year ago. Something more sophisticated than trial and error that really produces reliable results. Sorrowly, I lack both programming and super metroid skills, to get this done.
Not to speak of the time, because I am currently writing my own master thesis...

Smokey

2008-06-14 01:47:21 pm

I like horscht's method in general but I think its too complicated for any% runs considering its pretty short, but it definitely works on 100% runs.