No announcement yet.

Dustforce DX: "Tunnels skip"

  • Filter
  • Time
  • Show
Clear All
new posts

  • Dustforce DX: "Tunnels skip"

    Just incase this gets lost or deleted in the world of reddit here:
    Significant skips are rather rare in Dustforce due to the linear nature of levels and game progression (levels are segmented in an overworld, meaning a skip would have to occur within a ~25 second IL). While occasionally a new route might be found, and small optimizations are found all over the place rather often... it is VERY rare for a useful out of bounds clip or major "sequence break" to be found that completely redefines how a level is played. However, late at night this saturday, I found a new method of doing a long theorized skip, and with the help of others we managed to make it viable outside of TAS!

    Tunnels is a rather standard level for Dustforce, as can be seen by the current RTA record - it's very horizontal, boost oriented, and very linear. However, it did have a rather intriguing bit of geometry on a wall that when climbed, would push the player out of bounds. However, getting anywhere useful from here was a major uphill battle, as the next place useful to clip back in bounds was extremely high above the player. Here's an image of the area directly above this clip.
    Several years ago, a player who was known for finding absurd, broken routes (Ravencoff) found that if you used a super attack to refresh your aircharges, you could just BARELY reach the next "floor" and would skip about half of the level. However, achieving this super attack quickly was completely impossible - the only method for earning this super was to climb up to an apple, attack it 100 times (!), and then climb through the clip and use that super. Ultimately, the idea was a great meme, but it was obviously too slow and was forgotten about.

    More on apples, however - apples are easter eggs added by the developers that are supposed to be fun objects for adventurous players to discover. They are invincible, can be knocked around, and ultimately don't do much else. Except... if you repeatedly attack an apple in the air, you create a sort of "flying machine" where you push the apple around through the air, while the hitrise from attacking the apple causes you to slowly rise and float through the air. While this lets you cross infinite gaps both horizontally and vertically, it is an extremely slow form of movement (ESPECIALLY vertically) and it's just too impractical in most cases.
    Some players suggested (jokingly) to pair up this "flying machine" technique with the out of bounds clip in Ravencoff's route - instead of farming a super, use the apple to hover up (slowly...) to the next section of the level while staying out of bounds, and then get back in bounds and finish the level as normal. While technically this route did work, there was a few problems... for starters, the apple is about 6 seconds out of the way. As well, you would also need to force this apple to go into the clip while also keeping both of your aircharges on Dustkid, which is a very unwieldy feat and is also quite slow. After about 9 seconds of this setup, you're ready to (slowly...) hover up to the next part of the level. Even after you clip back in, you still half to play the rest of the level as Dustkid, who is rather poor at boosting in general, and would have to deal with finishing the end of the level without a super attack.

    The community assumed this surely would not be faster, and since at the time we lacked the ability to make optimized TASes or even really tool assisted demos, the route became an obscure meme in the community and was forgotten, and testing it seemed like a waste of time. Enter myself 3 years later, a person who has too much time on their hands and tests everything for no reason. Being "retired" from ILs and playing the game competitively, I now just spend time TASing the game or testing out cool routes or other ideas, often taking suggestions on dumb strats to test. I did some rough timings of parts of the route, and found that it was maybe possible for an optimized TAS to about match the current real time record, and made a very rough segmented demonstration of the route to test this theory. And, well... this low optimized demo saved about a second over the current TAS, and beat the RTA record by over 2 seconds.

    While people were surprised that this route was faster, it was far from human viable. For starters, the out of bounds movement is COMPLETELY blind. Therefore, you must carefully control your height, an apple's height and distance from you, and frequently change directions of your lights to speed up or slow down the apple - none of these adjustments could be made at all in real time. As well, getting yourself AND the apple clipped out of bounds in the first place is a fucking nightmare. You have to keep both of your aircharges, make sure both you and the apple are at the perfect height, and then do a bunch of uplights in a row to force the apple out of bounds - a number of uplights that could vary, resulting in a basically non-deterministic period of time it took for the apple to clip out of bounds, meaning you had to guess when to start the out of bounds movement and just hope the apple was in the right spot. After all this, you still had the ending to worry about, which rode on a very tight cycle and featured a boost that is essentially impossible for a human to perform in real time, which combined would save nearly a second. If you do the math, not being able to do this ending strat meant that you only had about a second of leniency to execute this insanely unpredictable and difficult route. I was very adamant about it being impossible in it's current form, which sparked interest in people looking in to it to make it more real time viable.

    Freshmaniac showed me a variation of the route that would instead push the apple to the left at the start of the out of bounds section, and then would clip back into the level at a point further to the left. Not only did this make positioning the apple relative to the player much easier, it solved some of the woes of having to adjust light directions to move the apple... it also was about half a second faster. Other players also suggested a few optimizations, and so I made a more optimized TAS that managed to save 3 and a half seconds over the old route TAS record, and about 6 seconds over real time - the margin of error seemed a lot more doable! With all of this, Fresh and I labbed out the difficult sections with save states, and managed to successfully find a 100% consistent way of doing the out of bounds section, as well as figuring out ways to make the starting section less terrible... and honestly, we were pretty unsuccessful in making this section not absolutely awful. It rides on a frame perfect input to put the apple at the right spot in front of the clip, and then the uplights to force the apple into the clip is still rather variant and inconsistent... but still doable for a human with enough patience! Fresh created this "human theory" segmented run for this route using these more consistent strats, and a couple hours later I managed this run that choked the last half of the level, which should be the easy part of the level.

    TL;DR: New skip on tunnels that will save about a second in real time, but is still extremely difficult and inconsistent - probably the most difficult strat in the game, currently. People have gotten close, and I expect someone to finish a run faster than the record within the week. To conclude, here's the current TAS record just to show the hilarious potential of this route.

    Considered making this into a video but was short on time, hope you like walls of text! Can answer any questions you have about the game or the route.