1 2
Posted: 3/28/2017 12:29 PM
Quote
blackisto
He/Him
Joined: 3/28/2017
Posts: 10
Thanks for your input. This "new" FBA is the one that's on this link, right ? (I put the link for reference) http://www.fbalpha.com/downloads/ I have also read about recent new versions of mame-rr. But I'll start with FBA, since indeed some work has been done on DDP-DaiOuJou (btw; I expect this game to be much harder to TAS than DDP !).
Posted: 3/28/2017 10:52 PM
Quote
Patashu
He/Him
Joined: 10/2/2005
Posts: 4045
I think so, more information here: http://tasvideos.org/forum/viewtopic.php?p=443235#443235 If FBA doesn't work for your TASing purposes, try FBA Shuffle.
My Chiptune music, made in Famitracker: http://soundcloud.com/patashu My twitch. I stream mostly shmups & rhythm games http://twitch.tv/patashu My youtube, again shmups and rhythm games and misc stuff: http://youtube.com/user/patashu
Posted: 3/29/2017 7:06 AM
Quote
blackisto
He/Him
Joined: 3/28/2017
Posts: 10
Yeah, with the FBA I got with my link I didn't find any frame advance feature, nor memory watch ---- which I found now with FBA-rr. I managed to use the frame advance feature with FBA-rr; and thus will start digging my first major question on stage 1-5 full chain. I will read your link again. It seems that frame advance is still possible with FBA...
Posted: 6/1/2017 2:00 PM
Quote
blackisto
He/Him
Joined: 3/28/2017
Posts: 10
Ok, here is a small update. I believe I now have a reasonable understanding of the different TAS tools -- while not being an expert of course. Regarding dodonpachi; a nice emulator could have been the recent mame-rr emulator. This emulator has been used to produce some TAS for other arcade games. However, for some reasons; save states and load states are not working on Dodonpachi. Developers seem to be working on some integration to Bizhawk (if I understand it correctly), which is great. Currently, for my use, FBA-rr remains the best solution. I did not try to record movies though (I'm not looking to submit anything at the moment); but frame advance works well; save/load states works well and RAM search / RAM watch work well; although the game is slow when the search is done on all the RAM (not an issue when you start filtering the RAM, when you look for something). About my four priorities (see, my previous post); - I've been able to identify some variables in the RAM which are responsible for the pattern of the 1-1 midboss. Of course; the 1-1 midboss is easy to beat; but i'm interested in RNG manipulation in general since this would be very useful for live runs, specially on the 2-3 boss. So I got a slot in the RAM. However; I've been unable to manipulate the values of these variables in order to consistently get a fixed pattern. The values change (very quickly) with the movements and with time. That's somehow a very bad news for live runs; but maybe somebody smarter than me will find a way. - I worked on the bullet cancel scoring mechanics; and now have understood most of it. I will post an article somewhere. The results I got are somehow unexpected. Most players believed that ; in stage 3; your score was growing exponentially with the number of bullet cancelled. That's not the case. I'll link the article here when it's out. :) - I didn't work on the 1-5 full chain (see my previous post); but I'm rather optimistic FBA-rr will help me to get a solution. Frame advance + save/load states should be sufficient. - I am nowhere close to finding the "100 million" magic trick that the world-record holder, WTN, found (and kept secret) in 2015. I recall that his world record is now 862 millions (was 755 millions before). With the knowledge we have; a very good TAS would yield around 800 millions. Knowledge of the game is missing here :(
Posted: 6/1/2017 8:16 PM
Quote
Patashu
He/Him
Joined: 10/2/2005
Posts: 4045
As for the RNG - I know with reasonable certainty that in Dodonpachi DaiOuJou, the RNG is advanced whenever anything 'random' happens, because if I take a savestate before a boss and use the laser a different amount, I get a different pattern, and I tend to see the same patterns over and over related to different lasering amounts. Any idea if something similar is going on in Dodonpachi? Or is it an input-reading RNG?
My Chiptune music, made in Famitracker: http://soundcloud.com/patashu My twitch. I stream mostly shmups & rhythm games http://twitch.tv/patashu My youtube, again shmups and rhythm games and misc stuff: http://youtube.com/user/patashu
Posted: 6/2/2017 9:08 PM
Quote
blackisto
He/Him
Joined: 3/28/2017
Posts: 10
The RNG for the patterns does not work similarly for all enemies with random patterns; but for the 1-1 (or 2-1) midboss; the variables are "frozen" approximately 2 to 3 seconds before the pattern begins; meaning that if you make a save state after that you will always get the same pattern no matter what you do (movements, laser, shot, etc.). The same happens for the 1-3 and 2-3 boss; except that the RNG is fixed approx. 1 second before the attack starts. Now before the RNG is fixed; the RNG depends on the movement but also on whether you use your autoshot, your shot or your laser.
Posted: 6/6/2017 10:59 PM
Quote
moozooh
Senior Moderator
Joined: 8/4/2005
Posts: 5777
Location: Away
blackisto wrote:
- I didn't work on the 1-5 full chain (see my previous post); but I'm rather optimistic FBA-rr will help me to get a solution. Frame advance + save/load states should be sufficient. - I am nowhere close to finding the "100 million" magic trick that the world-record holder, WTN, found (and kept secret) in 2015. I recall that his world record is now 862 millions (was 755 millions before). With the knowledge we have; a very good TAS would yield around 800 millions. Knowledge of the game is missing here :(
Hey there, thanks for taking interest in TASing the game. Please keep me informed of your progress (via PMs or otherwise); I'll likely be of use. Also refer to my earlier post in the thread, esp. points 5, 6, and 11. Most of what I said in 2011 remains valid. I'd also suggest keeping in touch with xy2_ (especially since you're both French, unless I'm misremembering something)—he has been working on a DDP DOJ TAS which has some similar tricks, like those related to laser wobbling. Cross-pollination of techniques is possible and would be nice. As for the 100 million trick, there are several theories. One is that there is a bonus for carrying the max bomb bonus from the start of the game all the way through Hibachi (I haven't checked it myself). Another is that there is a glitch/exploit, which, if true, you're very unlikely to discover on your own. Yet another theory is that WTN and the others who have updated their scores recently are playing all stages up to 2-3 (?) underpowered which lets them chain sections previously thought unchainable (most notably 2-3, but possibly 1-3 as well!). Keep in mind that the previous record was by no means perfect, and with 1-5 full-chain in mind the potential score for A-L had been considered to be in the whereabouts of 780–790 million, which means that only 70–80 million at most are actually "new" assuming the rest of the execution is better than in the previous record.
Warp wrote:
Edit: I think I understand now: It's my avatar, isn't it? It makes me look angry.
Posted: 5/24/2018 12:57 PM
(Edited: 5/24/2018 5:49 PM)
Quote
blackisto
He/Him
Joined: 3/28/2017
Posts: 10
Alright, after one year of important DDP playing (don't hesitate to say hi on twitch) and information gathering, I would like to post some updates: - It is confirmed that the new 862 millions World Record is some kind of bug. The old World Record was 755 millions and the new run is essentially as "good" as the old one, with an extra ~110 millions obtained from the bug. - forget about playing underpowered, or about carying the maximum bonus all along (this was already the case, by the way), there is one European player who managed to trigger a bug which generated an extra 15 millions out of nowhere. The player is Bananamatic and has, by the way, the community record at DDP (best score outside asia), which is currently 608 Millions. - The goal of this post is to share all the things I know because it would be great to have more people investigating for this bug. So here is what I know : 1) Bananamatic triggered the bug in stage 1-6 (i.e. first loop, stage 6) while he was training, using mame, with savestates. Unfortunately, he did not record his training; nor does he have an .inp file to share because you can't record an .inp file from a loadstate. It is thus not possible to replay what he did. I can only detail what he told me. 2) The route used by Bananamatic in 1-6 is essentially the chaining route of this video: https://www.youtube.com/watch?v=Ovf6xpH4dRE&t=1028 Bananamatic remembers he did a full chain (from the beggining, till the end of the stage) of approximately 809 hits (which is quite standard, and in fact quite low for a full chain). He did NOT use any bomb. The savestate he uses starts at 1-6 with a score of 127millions. After the chain is over (so before the boss of stage 1-6), he noticed he had a score of 208 millions, which is ~15 millions more than what he's supposed to have. We thus know that the bug is not triggered by bombing, and is not a bug on the boss. 3) From another source, which knows what the bug is (but cannot share it for reasons I won't discuss here) the bug is hard to find; but is not something completely gamebreaking which ruins the scoring of the game for people interested in superplays. Given these informations, I'm now sharing my thoughts: - stage 1-6 has some unique enemies which never appear in other stages, namely those small turrets which fire small blue lasers and which respawn. - I now believe the bug lies in calculation of the points you get from your chaining. Namely the "GP", sometimes called "base points" of some enemies. In short, each enemy is supposed to give some "base point". When you do a chain, whenever you kill an enemy, you obtain the base point of the enemy you killed plus the base points of all the enemy killed before in the same chain. So if b_n is the base point of the enemy for the n^th hit, this hit gives you b_1 + b_2 + ... + b_n points. Now if for some reason, you manage to glitch the base point of an enemy; the extra points you get will clearly snowball. For example, in a chain of 800 hit, if b_1 is normally equal to 100, but you manage to glitch b_1 to be equal to 10'000, then you will get an extra 800 * 9'900 points, which is a bit less than 8 Millions. If you manage to glitch more than one enemy, you would get more... I am not aware of any technique to glitch the base points of an enemy, but here TASing might prove very useful. - Given the relatively low hit count (~800 hit) that Bananamatic did, we have confirmation that the glitch does not come from some crazy, glitched, hit count. - To come back to the base points of each enemy, I would draw your attention on whether (a) the base point can be manipulated by changing the way you kill the enemy (with the shot, the laser, the aura, point blanking) (b) the base point can be manipulated by killing more than one enemy at a time. We already know that when you do a chain, the hit count can be slighty increased by lasering two enemies at a time. So maybe the game does not handle well multiple simultaneous kills. So there you go. Sorry for the tl;dr. I strongly believe that a careful analysis of this stage 1-6 using TASing tools would help to solve the mystery. Of course, I plan to work on it. Please don't hesitate to do the same and share your thoughts.
Posted: 5/24/2018 2:58 PM
Quote
moozooh
Senior Moderator
Joined: 8/4/2005
Posts: 5777
Location: Away
blackisto wrote:
- It is confirmed that the new 862 millions World Record is some kind of bug. The old World Record was 755 millions and the new run is essentially as "good" as the old one, with an extra ~110 millions obtained from the bug.
That's disappointing, even if not unexpected. It also potentially means one could possibly trigger it in other places once the exact mechanics are known, so basically any score targets are off for a proper TAS until it's fully researched. That said, it should still be possible to reach some 790–800 million points not counting the extra score from the bug.
blackisto wrote:
- forget about playing underpowered
For triggering the score bug, maybe, but a TAS doesn't have to forget that; that is, local techniques do not in any way invalidate the rest. You could still score extra on certain stages this way. It might also be the only way to full-chain stage 1-5 (at least in unassisted conditions), which I don't think has been solved by the Western community yet. Given the complexity of DDP and stage 3 design in particular, there might be quite a few loose millions to be discovered by applying a TAS-exclusive approach in addition to the score bug.
blackisto wrote:
carying the maximum bonus all along (this was already the case, by the way)
Do you mean WTN has actually no-missed/no-bombed through Hibachi? I don't believe that. His previous score would've been over 760M if that were the case, just from the normal life/bomb bonuses.
blackisto wrote:
(b) the base point can be manipulated by killing more than one enemy at a time. We already know that when you do a chain, the hit count can be slighty increased by lasering two enemies at a time. So maybe the game does not handle well multiple simultaneous kills.
This seems very likely. The basic hypothesis that comes to mind is that a value from one (expensive) enemy or destructible object is assigned to a much cheaper enemy/object. So either killing or spawning a cheap object at the same frame as killing/spawning an expensive one should prove or disprove this hypothesis. Since most enemies actually spawn offscreen in an invulnerable state, this might indeed be exclusive to enemies that only spawn inside the scrollable area: the turrets in stage 4 and 6, Hibachi clones in stage 6, possibly some others. It could be a combination of two enemies that spawn in the scrollable area, e.g. the Hibachi clones spawn near the respawning turrets, so both could be involved in the glitch. I would also suggest focusing on testing setups that can be sustainably reproduced in unassisted conditions, because I'm not 100% convinced WTN would bother with a very low-probability bug in a game as long and hard as DDP where the first instance of its application is about 20 minutes into the run.
Warp wrote:
Edit: I think I understand now: It's my avatar, isn't it? It makes me look angry.
Posted: 5/24/2018 6:08 PM
Quote
blackisto
He/Him
Joined: 3/28/2017
Posts: 10
Thanks for your reply Moozooh. I was hoping that you would reply with your thoughts, and didn't wait for long :)
moozooh wrote:
That's disappointing, even if not unexpected. It also potentially means one could possibly trigger it in other places once the exact mechanics are known, so basically any score targets are off for a proper TAS until it's fully researched. That said, it should still be possible to reach some 790–800 million points not counting the extra score from the bug.
800 millions seems indeed possible in a proper TAS, without the bug.
moozooh wrote:
For triggering the score bug, maybe, but a TAS doesn't have to forget that; that is, local techniques do not in any way invalidate the rest. You could still score extra on certain stages this way.
Sure. I was indeed talking only about the bug. For a TAS, this is a different story, although I doubt that it would work because playing underpowered slows down the pace at which the hits increase when you laser a big enemy.
moozooh wrote:
It might also be the only way to full-chain stage 1-5 (at least in unassisted conditions), which I don't think has been solved by the Western community yet.
The setup to full chain stage 1-5 is now known. I managed to pull it alone using FBA-rr (with an unrealistic route for a live play); and the Japanese players found a way to get it in live runs. There is no need to be underpowered.
moozooh wrote:
Given the complexity of DDP and stage 3 design in particular, there might be quite a few loose millions to be discovered by applying a TAS-exclusive approach in addition to the score bug.
Agree. For example, a chain could be maintained from the stage 3 midboss, till the end of stage 3, while optimizing the bullet cancel section.
moozooh wrote:
Do you mean WTN has actually no-missed/no-bombed through Hibachi? I don't believe that. His previous score would've been over 760M if that were the case, just from the normal life/bomb bonuses.
The maximum bonus no longer gives any point after the 2-6 boss. The WR is essentially a run with no miss no bomb until the 2-6 boss; then no miss (but maybe some bombs) on the 2-6 boss to get the 8 millions NoMiss bonus; then 1 miss on the giant bee+hibachi. Without the bug, WTN believes that his theoretical maximum (with his strategies and a full nomiss) is 790 millions.
moozooh wrote:
This seems very likely. The basic hypothesis that comes to mind is that a value from one (expensive) enemy or destructible object is assigned to a much cheaper enemy/object. So either killing or spawning a cheap object at the same frame as killing/spawning an expensive one should prove or disprove this hypothesis. Since most enemies actually spawn offscreen in an invulnerable state, this might indeed be exclusive to enemies that only spawn inside the scrollable area: the turrets in stage 4 and 6, Hibachi clones in stage 6, possibly some others. It could be a combination of two enemies that spawn in the scrollable area, e.g. the Hibachi clones spawn near the respawning turrets, so both could be involved in the glitch.
That is very interesting and worth testing. I like the idea of small enemies on the scrolable area which would take the value of big enemies.
moozooh wrote:
I would also suggest focusing on testing setups that can be sustainably reproduced in unassisted conditions, because I'm not 100% convinced WTN would bother with a very low-probability bug in a game as long and hard as DDP where the first instance of its application is about 20 minutes into the run.
Yes clearly. Even if he's WTN, he doesn't like depending on RNG or stuff like that. By the way, WTN plays time A; and we know that the type C WR (which belongs to CLOVER TAC) also uses this glitch. Same for the type B WR.
Posted: 5/24/2018 6:40 PM
Quote
moozooh
Senior Moderator
Joined: 8/4/2005
Posts: 5777
Location: Away
blackisto wrote:
The setup to full chain stage 1-5 is now known. I managed to pull it alone using FBA-rr (with an unrealistic route for a live play); and the Japanese players found a way to get it in live runs.
Oh wow, now that is some good news! Do you have a video of that anywhere?
blackisto wrote:
The maximum bonus no longer gives any point after the 2-6 boss.
The reason why I'm thinking this is that in DOJ you get a very substantial bonus for carrying the max bonus through Hibachi, but as I'm not very active in the community anymore I don't know if anyone has ever confirmed if there's a similar one for DDP. On a slightly unrelated note, have you been to Stunfest this year? I heard fufufu played DOJ WL there for the big screen.
Warp wrote:
Edit: I think I understand now: It's my avatar, isn't it? It makes me look angry.
Posted: 5/24/2018 10:26 PM
Quote
Patashu
He/Him
Joined: 10/2/2005
Posts: 4045
This is a very cool writeup! Looking forward to this glitch being figured out. Agreed that ‘same frame confusion’ style glitches are going to be the most likely cause. I’ve actually been playing DDP lately, practicing for a scrubby 1-ALL, but I’m no where near good enough about the game to test scoring related hypotheses, so I’ll be cheering from the sidelines. Some thoughts: -If Dodonpachi has sprite slots ala SMW and SM64, one of the glitches could depend on which slots are filled with what. -Actions that could happen simultaneously to trigger the glitch: Enemy killed by shot; enemy killed by laser; enemy killed by laser aura; enemy spawns off screen; enemy spawns on screen; enemy stops being invulnerable. And all of these could depend on slot numbers if slots exist. Good thing we know it’s not related to bombing, dying or dropping chain. Picking up a powerup, bomb or star probably doesn’t matter unless they use the same slots as enemies and slots matter. Bullets use their own system so they should totally not matter. Could uncovering/picking up a bee matter at all? -Your best shot is probably to try different hypothesis like 'if I do X and Y on the same frame instead of on separate frames, do I earn more points?' for every combination that could come up in 1-6 practice, like you said. There's likely some generic application that can be used in most stages, though that's just a guess - if it's related to the respawning turrets, then it's only possible in stages 4 and 6, for example.
My Chiptune music, made in Famitracker: http://soundcloud.com/patashu My twitch. I stream mostly shmups & rhythm games http://twitch.tv/patashu My youtube, again shmups and rhythm games and misc stuff: http://youtube.com/user/patashu
Posted: 5/24/2018 11:59 PM
Quote
moozooh
Senior Moderator
Joined: 8/4/2005
Posts: 5777
Location: Away
Note that the turrets in stages 4 and 6 are different; the ones in stage 4 spawn once and leave a ground star on destruction while the ones in stage 6 respawn continuously until scrolled to the bottom part of the screen. This may or may not be relevant. Both spawn within the scrollable area, however.
Warp wrote:
Edit: I think I understand now: It's my avatar, isn't it? It makes me look angry.
Posted: 5/25/2018 7:46 AM
Quote
blackisto
He/Him
Joined: 3/28/2017
Posts: 10
moozooh wrote:
blackisto wrote:
The setup to full chain stage 1-5 is now known. I managed to pull it alone using FBA-rr (with an unrealistic route for a live play); and the Japanese players found a way to get it in live runs.
Oh wow, now that is some good news! Do you have a video of that anywhere?
blackisto wrote:
The maximum bonus no longer gives any point after the 2-6 boss.
The reason why I'm thinking this is that in DOJ you get a very substantial bonus for carrying the max bonus through Hibachi, but as I'm not very active in the community anymore I don't know if anyone has ever confirmed if there's a similar one for DDP. On a slightly unrelated note, have you been to Stunfest this year? I heard fufufu played DOJ WL there for the big screen.
I do have a video, sent privately by WTN. I will not publicly post it but I can send you a link by PM. In addition, I will gladly share it with anybody who is willing to TAS this game or who is willing to do the 1-5 full chain in a live run (but good luck ... ;-) ) Regarding the maximum bonus, there is no such mechanism in DDP. All the points you get from the max bonus are obtained once you reach the 2-6 boss. Then you can bomb/die etc; and it will not change anything to the maximum bonus. It is still important to survive because of the NoMiss bonus on 2-6 (8 millions) and the NoMiss bonus on Hibachi (but good luck for this one in a live run), and the 10 Million point bonus per remaining lives once you clear the game. In any case, you have the right to bomb on the 2-6 boss/ the giant bee/ Hibachi, without being penalized. Oh, and yes I was at Stunfest. The whole STG crew was here and Fufufu's runs were absolutely stunning to watch. Replays are available if you go to the Stufest2 twitch channel.
Posted: 2/16/2020 12:55 PM
Quote
blackisto
He/Him
Joined: 3/28/2017
Posts: 10
EDIT : The glitch has been found !!! will be made fully public shortly (give us some time to gather all pieces of evidence :) ) https://imgur.com/a/8ez3MHo We are writting a complete article. This will be VERY interesting for TASers because, in certain cases, the activation of the glitch is very hard even when you TAS !
Posted: 3/1/2020 5:42 AM
Quote
Patashu
He/Him
Joined: 10/2/2005
Posts: 4045
The Dodonpachi glitch has been documented and revealed here: http://electricunderground.io/the-secret-dodonpachi-scoring-glitch-revealed/ tl;dr it's a frame perfect glitch that doubles how much your combo is worth and can be done 4 times in stage 6. Is it possible to counterstop the game in TAS conditions now...? Possibly!
My Chiptune music, made in Famitracker: http://soundcloud.com/patashu My twitch. I stream mostly shmups & rhythm games http://twitch.tv/patashu My youtube, again shmups and rhythm games and misc stuff: http://youtube.com/user/patashu
Posted: 3/1/2020 2:01 PM
Quote
blackisto
He/Him
Joined: 3/28/2017
Posts: 10
Thanks Patashu :) The link given by Patashu explains our findings with the glitch. I also tried to activate it in live play (but with a bunch of save/load states !) https://www.youtube.com/watch?v=_C4a9W0hIp8 Now, all the ingredients are here to do a nice TAS of this game. Except maybe a decent TASing tool ? I don't know much if things have evolved. FBA-rr is really... not good is it ? In a TAS, one may expect to get a glitchless loop1 at ~225-228 millions including a 1-5 full chain -- a setup is now given in this video: https://www.youtube.com/watch?v=HLJjsSngYFk So this is glitchless. If you add the effect of the glitch, you may add approx 120 millions if you do a 'triple glitcho', and probably around 250 millions if you manage to do a "quadruple glitcho" (i.e. activate the glitch on all the Glitcho enemies in 1-6). I need to point out that quadruple Glitcho is very challenging, even in a TAS ! Then in the second loop, same story. Triple or Quadruple Glitcho in 2-6... Finally the gain from the glitch goes from 240 millions (triple glitcho in 1-6 and 2-6) to 500 millions (quadruple glitcho in 1-6 and 2-6) which means counterstop is possible in any case. Does anyone want to TAS this game ? That would be hype. I'd be glad to help regarding routes/what you may expect in terms of hits / score. But I am not a TASer...
1 2