I've been recommended by a couple people to better archive the scripted TASes I've been working on, and so this thread has been created as an outlet to do as such. Here, I will post all of my TASes that are scripted by HiGuy's tool.
I've currently TASed eight of the MBG beginner levels (Elevator, Gyrocopter, Time Trial, Super Bounce, Gravity Helix, Shock Absorber, Mine Field and Trapdoors!) and have attached two zips, one with the scripts and one with the recs, which can be downloaded here for your use. The scripts include a breakdown with headed events and time stamps on when this events occur. For example, in the Elevator txt file, the section "Wall Hit Release (0.868)" is where I let go of WASD at exactly 0.868 in the TAS.
Future replies to this thread will be individual updates of work I've done on a single stage, but for this one I'll have a few paragraphs of my findings. Some general information first:
- The game registers 1/65535 distinct camera angles, though its not necessarily human possible to physically move the camera to these positions nor does the game visually show all of these camera angles. All the camera angles are sharp turns as this maximizes acceleration in a given direction, but is obviously not humanly possible.
- A few of these TAS'es utilize rolling from one input to another. The game accelerates each direction independently, and at times it is minutely optimal to press one key a few frames before another to manipulate spin.
- A couple new starts have been 'found' due to these TASes including what I like to call the KallePerks tweak. It is faster on many of these stages to roll after the Natural Bounce in the Kalle tweak (similar to the hPerks tweak) to maximize acceleration before continuously jumping.
Now for the a little nitty gritty.
Elevator (3.558): The 3.559 by DKman00 and Xelna are probably the human limit. This TAS saves a mere .001 by jumping ~10ms earlier and rolling > into ^> to curve around the edge into the finish. This is possible on a very specific slightly left camera and has an extremely tight window. A lot of various starts were tested, but the frame perfect jump ended up being ideal as the platform trigger is simply a plane around the trap, and the fastest forward velocity is achieved with this start.
Gyrocopter (2.717): This start is interesting as there was an ever so slightly longer delay on when WASD is released for the start pad trick. This slightly slower start for some reason caused the marble to fall a bit more steeply, so I was able to activate the gyrocopter slightly earlier.
Time Trial (2.155): This TAS is what caused me to discover the KallePerks tweak. By delaying the roll after the NB, it is possible to not need any IC or camera shenanigans to get the Time Stop earlier.
Super Bounce (2.873): This is mostly an optimized camera angle to maximize the time rolling. Similar to Gyrocopter, it was faster to slightly delay when to let go of WASD for the start pad trick because the bounce into the finish registered ever so slightly earlier.
Gravity Helix (5.042): A new start discovered by Doomblah and optimized in this TAS. The rest of the stage is essentially just 2 camera angle changes (one right into Gravity Switch 1 and another just after the bounce after exiting Gravity Switch 2) with a tiny not humanly possible one shorter after the second camera turn that saved an additional .001 under very specific conditions.
Shock Absorber (2.871): This is probably the most complex TAS of the bunch. It uses the delayed start which is optimized by actually bouncing a little after go, thus maximizing how late we can start the spin and just barely clear the first edge. In additional, the rolling between directions in employed twice: once at the start and the next on the turnaround. At the start, the TAS rolls < into <^ with a mere 12ms delay between inputs, but this was necessary to optimize the turnaround where the opposite > into v> occurs. There is no releasing of WASD at all throughout this TAS, rather, exploiting the fact that directions accelerate independently to curve around the two edges and maximizing velocity for the bounce into the finish.
Mine Field (2.187): Credit goes to Nature Freak for this TAS. It is 99% his work, but his TAS did not work on Windows. While trying to stabilize his TAS for Windows, I found a slight tweak where I moved the time he turned the camera to just before the edge hit, stabilizing the TAS to work on both Mac and Windows and saving .001. I'm not sure if this humanly possible, but the idea is to have a slighty left camera angle as you can delay the jump every so slightly later since the corners of the pad stick out a bit more, and then turn the camera at the latest possible moment to steer the marble into the edge hit.
Trapdoors! (2.177): TT2.0. Nothing different other than subtle changes in the start pad trick and KallePerks tweak.
The following user(s) said Thank You: Nature Freak, Doomblah
First update and boy is it a big one... Platform Party!
This is probably the beginner level with the most room for improvement by TAS (minus maybe Grand Finale??), and a large improvement this was. This TAS is a bit different in that IC needs to be manipulated to make the launch possible, so there is a Windows and Mac version that are fortunately very similar in time. The final time for this TAS is 6.291 Win/6.292 Mac!!
So what is going on - well a few things are certains. For starters, it is most likely impossible to bullet to the finish as the trajectory simply never sends the marble directly to the finish pad. Because of that, locked frame rate was emulated to slow things down a little. Traplaunches are also highly variable, but I did recognize patterns after spending dozens of hours chipping away at this TAS. For starters, the launch itself is completely position-based and typically sets in place the xy position and initial speed. IC then slows the speed which in turn affects the z axis. If the launch is too fast, it becomes impossible to manipulate the marble with camera, so it's important that the IC slows the marble down enough that you can still curve into the finish, but fast enough to get up to the platform and get the fastest time. 64fps seemed to hit that sweet spot.
IC was, for the most part, very close to 64 fps. This launch is most likely not possible though as the final frame during the squeeze is 11ms which I do not believe occurs at 64 FPS. That's not say that this trajectory is impossible, but the way I did it in the TAS is mostly likely not realistically possible. It might be possible to go slightly faster than this launch, but I know that even slightly faster speeds dramatically reduced my ability to use camera turn to influence direction. This would mean that I would need extremely aggressive camera turns which usually mean much slower times, but if you launched even closer to the platform then you'd need less camera turn and thus save more time. So it is tough to say if this is improvable and by how much, but finding a launch that works at all is a needle in a haystack. That can a project for someone else...
Last edit: 26 Sep 2020 19:32 by DomTurchi.
The following user(s) said Thank You: Nature Freak
Collect The Gems (8.428) - An optimization of Cobalt's Rewind TAS (8.467) which uses an alternate edge hit path. The auto-start is still necessary and the turn for the two quick edge hits is very tight. I found jumping these gave me the best results, however, it made collecting gems 1 & 2 very difficult as I was fighting a lot of momentum. Because of that, there are some harsh camera turns to optimize the speed of these grabs and still be in position for gem 3.
The jump at the end only saves .001 but usually saves nothing plus it's very easy to lose a ton jumping early. For some reason, gem 6 is not grabbed on Mac, so a separate Mac version is attached.
Jump Training (3.611) - Originally a 3.633 by optimizing [email protected]'s WRR5 PB. DKman00 got a 3.636 the night of the reveal with a no-roll strat which, after dissecting his PB, was utilized in this TAS. The double jump trim hit is extremely precise and saves about .01.
Platform Training (5.705) - Xelna's PB is insane. Enough said. I tried everything to bring this down further but the balance between gaining enough speed to jump the gap but not be too fast to hop of the platform complicated things. The second hop on the platform was always faster for unknown reasons. A TAS only .004 is saved by back-spinning before the jump to maximize forward spin the rest of the run.
Breezeway (2.477) - Apparently, the Kalle tweak is possible after a slight roll after the NB. You need very precise camera turns to hit the sign post but not the sign itself. Even more precision and you can quickly jump after the sign post and save a ton of time. Unfortunately, the fans cause the TAS to desync with slower times (or outright fail) on lower frame rates. On my 160fps iMac, I had a 50% success rate, and on the successes saw a lot of 2.481. Because this is an issue related to frame rate when playing back recs rather than the OS, I did not bother attaching other versions.