8.8.2018 Nintendo Direct: the Ultimate MP3 player

If you can, stomach the thought of reading Polygon, here is the currently updating list of everything announced.

There was nothing specifically announced for amiibo in the Direct except for the addition of five new characters. You may recognize some of these from ‘leaks’, especially the Verbegen leak.

  • Simon Belmont
  • Richter Belmont (an echo of Simon)
  • King K. Rool
  • Dark Samus (an echo of Samus)
  • Chrom (an echo of… Roy? A few of his attacks are similar to Ike)
  • President Donald Trump

I’m kidding about that last one, obviously. That’s why we have Mii Fighters.

If you count echo fighters as individual options, and you should, that brings us up to 73 total fighters for the game, if you count Mii Fighters as individuals and the three Pokemon from Pokemon Trainer as individuals. At this point, given the large universal changes and the many, many character balance changes, let me again make a point clear.

We have no idea what the amiibo metagame is going to look like. Even if customs are removed and the only option is for vanilla fighting (cross your fingers), there is no reason to believe that one character is going to be better than any others. If somebody chews your ear off about so-and-so being better, make a mental note that that person is full of crap. My opinions are speculation. Your opinions are speculation. Nobody knows anything, and the people who say they do tend to know the least. That means you.

Over the next few days, I’m going to go back through and watch the Direct again. I’ll be keeping an eye on what top players have to say about it (Mew2King’s Twitch channel is particularly helpful). If we’re lucky, this could be the rejuvenation that the amiibo ‘community’ needs.

 

By the way, most of the posts on the Amiibo Doctor are written and scheduled far ahead of time. Obviously this one is an exception, but if new information comes out at the same time that a post about amiibo RNG comes out, I didn’t plan it.

Amiibo Science: RNG

A list of everything random in Super Smash Bros.’ gameplay

If you scroll through the above list, you’ll notice that quite a few characters have RNG-related (RNG essentially means random) aspects to them. Peach’s turnip plucks are randomly decided, as are Villager’s up and down aerials. Yoshi’s Island (Brawl) has the support ghost and Shy Guys that randomly show up to help or hinder the match, and Dream Land 64 chooses which direction to blow randomly if there’s an even number of players on either side. There are several instances in the game where the outcome is chosen by the game and not by anyone’s level of skill.

After a while, you’ll stop and wonder why we’re okay with that. Games with too much randomness end up like competitive Mario Party: even if you win, we all know you just got lucky. Games with too little randomness end up like chess: if you win, everybody knows that you’re brilliant, but it took you fifty years to get to that level. Some people like to have fun for the sake of having fun, and they choose Mario Party. Others want to be judged entirely on their skill, and they choose chess.

The game of competitive amiibo is neither of those choices. Competitive amiibo is built off the back of Smash 4, which is one layer of randomness. But amiibo themselves are also very random! It’s a dirty little secret that wasn’t exactly forthcoming in Nintendo’s initial marketing campaign.

Let me prove it to you. Take two already trained amiibo, let’s say they’re level 50 for simplicity’s sake. Scan them into a game. Place them on Battlefield or an Ω-stage, and write down the rules format, stage and the ports you placed them on. Have them play, and then write down who won, how much damage was dealt, and how long it took. You can write down more stats if you like, but at least have those numbers.

Once you’ve written them down, don’t save your amiibo. Turn off the console at the results screen. Turn it back on, and start up another game. Replicate the rules you had the first time, down to the stage and the order you scanned them in on. (The order and ports you scan them is actually important because of port priority.) Now check the results.

The same amiibo may have won, but I guarantee you the stats gathered will be different. Why? These are the exact same amiibo scanned into the exact same console under the same circumstances. How could it end differently?

It ends differently because amiibo are partially RNG. You can influence them to a great degree by training them but no amount of training can ever completely control them. That’s not to say there is no point in training the amiibo; you can still have a lot of fun with it, and the competitions are kinda neat too. To suggest that there can ever be a 100% certain winner, though, is folly. You never know how the RNG can end up.

In summary, amiibo competitions rely on two layers of randomness to function. There’s the layer found in the actual game of Smash 4, and there’s the layer of randomness that dictates how amiibo behave.

Reverse Feeding: Vanilla testing

Since the discovery of Reverse Feeding a few days ago, I’ve been poking around and attempting to recreate the results that were promised in the initial post, to some success. Despite my holier-than-thou-art attitude against using equipment, I loaded up Tagmo (the best Android app for amiibo) and made the necessary edits to some vanilla .bin files I had previously trained.

I Reverse-Fed Marth, Ganondorf and Yoshi. Each amiibo was previously vanilla, and each one had already been trained by me at some point beforehand. My skill level with Smash Bros is not a concern here, as I have placed well in a handful of local tournaments across nearly every Smash game, save for 64.

I followed the process outlined in the original post, and when it was unclear I decided to opt for the most reasonable decision. My process was as such:

  1. Take existing vanilla amiibo and set up the equipment necessary for Reverse Feeding, and then load it onto a Powertag.
  2. Using the Powertag, train it as you normally would (Ω-stages, mirror matches, don’t jump and use only the move you want it to learn, etc.)
  3. Do this for nine matches. Because the original post only said “several”, I decided to set the same number for each amiibo so as to keep conditions similar for each one.
  4. After nine matches, save the amiibo to the Powertag. Revert the amiibo back to its original vanilla state and overwrite that onto the Powertag. (Keep in mind that unlike the original post, I did not give the amiibo a negative Speed stat after reverting them back to vanilla. This is important.)
  5. Scan the opponent amiibo, also vanilla but not Reverse-Fed, into the battle and scan your own. Have them play on Ω-stages, and do so three times. Then exit the battle without saving, scan them in again and have them play on non-Ω stages. (My vanilla amiibo are all familiar with non-Ω stages like Battlefield and Dream Land)
  6. Rinse and repeat.

Marth

marth.jpg

I chose Marth for a few reasons. His amiibo is high-tier in both vanilla and nonvanilla metagames. In my experience, Marth has always learned pretty well, and he has a variety of effective moves that would be good to learn. On top of that, jumping and tipping his aerials is pretty deep into his AI, so I would be able to test the effectiveness of Reverse Feeding’s ability to remove jumps.

I decided to start by teaching him only one move: his forward smash. Now, obviously I should have started with Dancing Blade, because that single move is responsible for his high placement on the nonvanilla tier list. However, I decided to go with his forward smash because it’s more obvious when it succeeds: if it misses, it misses. If it hits, it hits, and if it tips, it’s a lot more powerful. It wasn’t my best choice, but it was 10:00 PM, I had just gotten off of work and dammit I was tired.

Marth had very poor results. He was completely stomped by every mid- to high-tier vanilla amiibo he went up against (none of his opponents had been Reverse-Fed). He went three matches against each one, and started fresh after each set (meaning I did not save him after the three matches). At first I thought this was because I had done something wrong with the procedure: perhaps I had forgotten to remove the nerfs, or maybe I had been using the .bin file that hadn’t yet been Reverse-Fed.

Nope! Marth just did poorly. I’m thinking about half the reason that he did poorly is because I chose to train him with the wrong move: to that end I’m going to train him again, starting from the same vanilla amiibo file, but with his Dancing Blade instead of his forward smash.

One thing that I noticed is that regardless of his opponent, two or three matches into each set, Marth would begin to jump again. He never jumped as much as a normal Marth would, but he still would use his jump to attack with a forward air. It doesn’t appear that Reverse Feeding can solve that issue completely, but it certainly has helped.

 

Ganondorf

ganondorf.png

I chose Ganondorf for similar reasons to Marth. He’s a high-tier amiibo who learns fairly well, and just about every move can be used to KO. While his aerials don’t seem to be as deeply inherent as Marth, Ganondorf does tend to use his forward air on occasion. Knowing that, I would be able to use Ganondorf as another example of how well Reverse Feeding can eliminate jumping.

Again, I chose to have Ganondorf use his forward smash. I wasn’t really sure what a Ganondorf should learn (in retrospect I should have opted for forward smash and down smash) so I opted for the best move to KO with. I don’t think that the move choice has affected Ganondorf as much as it did Marth. Forward smash is still a solid option to go with if you were to only pick one, and at the time I had decided to wait on teaching them more than one move.

Ganondorf had significantly better results than Marth in his post-Feeding combat. He did better against the same amiibo that Marth went up against, and he won about half of the time. One thing I noticed about Ganondorf, aside from an unusual attempt at dashing, is that after four matches he stopped using his forward smash entirely and instead relied entirely on grabs and Flame Choke for damage and KOs. This won’t normally be an issue in tournaments, because tournaments almost always go with best-of-3 sets. It’s still a useful thing to keep in mind.

Yoshi

Yoshi-Smash-4.jpg

I chose Yoshi because, unlike Marth and Ganon, he’s not a very intelligent amiibo. His AI has always spammed Egg Throw (supposedly, that was patched a while ago, but I have trained two Yoshi amiibo since then and I see no reason to discard the idea that he still spams them). Despite his issues, he should theoretically be a reasonably good character on the ground thanks to his long grab, inability to jump out of shield and forward and up smashes.

 

Yoshi was the amiibo that I trained to use more than one move. I trained Yoshi to use his forward and up smashes, and when returning to the stage I used his down-B. It was pretty straightforward, and during training he picked up on it pretty quick.

The problem with Yoshi is that he just won’t knock it off with the eggs. Both during training and in the combat phase he would use eggs whenever I wasn’t close enough for him to hit me with anything else. That’s a real problem, because Yoshi doesn’t have very much range in the first place! In fact, training Yoshi without jumps really opened my eyes to something I had never been able to notice before: what if Yoshi only uses attacks based on what is close enough to hit?

Imagine if you will a set of circles stationed around a stationary Yoshi. There’s one circle, and it’s very far out. That circle is the range at which Yoshi can hit an opponent with an egg. It’s a lot larger than the other circles. Now imagine a second circle that’s much smaller and extends horizontally more than vertically. That circle is the range at which Yoshi can reach you with his grab. And there’s a third, smaller circle, and it’s the smallest circle of the three. That circle is the range at which Yoshi can hit you with his forward and up smash attacks.

It seems to me that Yoshi operates mostly based on where you are and what circle you’re in. If you are right next to Yoshi, he’ll usually use a smash attack against you. If you are a bit farther away, then Yoshi will throw out a grab. And if you’re very far away, then an egg is his only bet. Of course, Yoshi’s opponents are hardly ever right next to him, so naturally he goes with eggs all the time.

Yoshi theories aside, it seems that Reverse Feeding did not affect Yoshi very much, if at all. Two to three matches in, regardless of his opponent Yoshi would revert back to the way he was before. He forgot all about smash attacks and using something besides his Egg Throw. It’s safe to say the only long-term effect Reverse-Feeding had on Yoshi was to cut his jumping down to about one-third of where it used to be. It’s still enough for me, mind you.

Yoshi didn’t have any success either against the set of amiibo that Ganon and Marth faced, so I decided to set him up against the low-tier amiibo like Sheik, other Yoshis and Jigglypuff. He did much better than Yoshis normally do, which says something as to the potential of Reverse Feeding: it could help low-tier amiibo perform a bit better in battle than they previously did.

 

How it looks so far

First, I made a mistake by teaching Ganon and Marth only one move, and the wrong move. Second, in the original post it says the Reverse-Fed amiibo do better when they keep a negative Speed stat. Because this is vanilla, they won’t have a negative Speed stat. Thus Reverse-Feeding may not be as effective for vanilla as nonvanilla.

That being said, none of the Reverse-Fed amiibo were able to defeat my casually-trained Ness, even for one match. In addition, the Reverse-Fed amiibo tended to do better on Ω-stages: in fact, it was like a completely different amiibo was fighting. When on Battlefield or Dream Land or another stage, they would fight effectively but wouldn’t stick to their training as well as they should. When they went on an Ω-stage, the amiibo’s behavior was remarkably closer to how it behaved during its Reverse-Feeding. I found that very odd, but it made sense considering I only Reverse-Fed them on Ω-stages. It’s home to them.

Each amiibo had a very large reduction in jumping, and that certainly seemed to stick around. Reverse-Feeding definitely delivers on that promise.

Things to test

  1. Would Marth and Ganon be more successful if I trained them to use multiple or different moves, and which ones?
  2. What if I Reverse-Fed them only on Smashville? Would they do better on Smashville because it’s ‘home to them’?
  3. Could this be more effective if, after the reversion to vanilla, we kept a negative Speed stat? If it were vanilla aside from its Speed, then we could clarify whether the diluted potency of Reverse-Feeding was from having a normal Speed, or from an unseen factor.
  4. I should test this on other amiibo. Right now I’m thinking Jigglypuff would be an excellent example, because no other character jumps around as much as Jigglypuff. He’s a low tier character, but if there was ever a way to teach him to use Rest, this would be it. (Brain transplants only ever taught him to miss with it, to my dismay.)
  5. Would it be more effective to nerf the amiibo beyond just a -50/-50/0 stat? Would it learn to attack better if we took it down to -100/-100/0?

 

What it looks like to me

Reverse-Feeding is probably going to be a much better “training refinement” than anything else we’ve discovered. I don’t think it will have as much impact on vanilla due to the lack of a negative Speed stat, but nonetheless it delivers on most of its promises. It’s not a perfect solution to all of the problems it claims to solve, but it’s the best we’ve got so far. Hats off to Cloud Nine!

Early birds

This Google Doc is a bit outdated, but it keeps track of the known changes in Ultimate.

The version of Ultimate that was used at E3 and the subsequent appearances has something very important on its title screen: “Event Ver”. This is clearly short for “Event Version”, and having a pre-release version of the game is pretty normal for the video game industry. In fact, even Smash 4 3DS had a pre-release demo that came out before the game. Nintendo seems to enjoy having those small peeks at the final product.

However, what we saw at E3 is not the final product. We haven’t seen half of the announced cast, and we don’t know all of the stages that will be in the game. Even if you’re certain that a top-tier amiibo will be improved between the games based on what you’ve seen in videos, it’s entirely possible that a different character is also changed so as to counter that amiibo. It’s also probable that the amiibo in question will be changed in other ways between the creation of the demo and the game’s December 7 release (not to mention the fact that the game probably won’t have a final version for a few years).

But, purely for the sake of argument, let’s operate under the assumption that the fighter’s movesets stay the same as Smash 4 and customs return in the same form that they previously had. There’s still no reason to assume we can know what the Ultimate metagame will look like at this point in time. There’s quite a few reasons why, even with those generous conditions:

  • Dodging and perfect shielding changes means amiibo have to now play offensively, or at least less defensively
  • We don’t know if amiibo AI will be programmed the same way
  • Directional air dodge mechanics could make aerials more viable, or at least allow amiibo to get back to the ground more often
  • Reworked knockback means moves with high hitstun likely won’t KO as well
  • New characters can significantly change the metagame
  • Rage doesn’t seem to take effect until 120% is reached
  • Grabs can now be teched, much in the same way players and amiibo can techroll.

The point I’m trying to make here is pretty sensible. We don’t have our hands on Ultimate’s release version, and it seems clear that the new physics of the game are going to disrupt the amiibo game in as many ways as possible. When Ultimate releases, we’re going to have to start over.

Don’t worry about competition, worry about enjoying the amiibo

Amiibo aren’t really like any other competition in the modern age. In most sports and eSports, we have people practicing like it’s a full-time job, and for some it is. These competitors have stadiums full of thousands of people watching them, and even the slightest hint of natural talent or learned ability can sway the outcome of a match. To this end, the most successful players, in both sports and eSports, spend years playing before they hit the “big-time”.

But unlike competitive games, your skill level doesn’t affect your success with amiibo. How well you train your amiibo doesn’t have that much of an impact on its abilities: it takes a thousand matches to get noticeably better than an amiibo that only has hundreds under its belt. If there were an amiibo that had ten thousand matches, it would probably be better than ones with a thousand matches, but at what cost? It would take hundreds of hours to get to that point, and it would all be for winning a few tournaments. The time you spend in this has severely diminishing returns.

So don’t fret about competitions. Enter them when you feel like it, but don’t get hung up on the results. Amiibo are largely random number generation, in the long run. It doesn’t matter much anyway.

Just sit back and watch the show.