Game objectives
- Emulator used: DeSmuME 0.9.10
- Aims for fastest time
- Uses hardest difficulty
- Manipulates AI
Tricks used
AI Manipulation
Here is a state diagram of the AI:
At its fastest, it places a piece every 23 frames. It only hard drops (no soft drops). And, importantly, once it has chosen a placement, it does not consider a better placement for the same piece. That leads to situations like this:
In the screenshot, the AI has already chosen the best placement, even though I will soon be adding pieces that will give it a far better placement:
The last time the AI reaches the decision/"hard drop" state is with an I piece.
Afterward, it finds no desirable placements,
gravity makes the AI's pieces fall and they are only placed due to
lock delay.
Run details
Here is my strategy for the run. The
Random Generator seed for Push Mode is always the same, provided it is the first game played on startup, even if a variable number of frames have passed until the game starts. Since Push Mode only rewards line clears (not T-spins or combos), my goal is perform 2 Tetrises as quickly as possible. Due to shortcomings in the level 5 Push AI, this is sufficient to win. I use "hold" twice.
As for tetromino efficiency, there is only 1 mino above my platform after the 2 tetrises are performed, so the number of pieces used (23) is optimal, I think. However, I do use "hold" twice, which costs 2 frames total. If someone found a strategy that places 23 pieces but uses no holds, it would be 2 frames faster than mine.
Similarly to ToonLucas22's TAS, the replay file ends several seconds before the Push Mode game itself has completed. This means that if a Push Mode TAS were to be combined with any other modes, the extra seconds would need to be included in the replay file, and the strategy used here would become very inefficient. For this reason, I do not recommend trying to include this in a TAS of several Tetris DS modes at once, as efficiency in the individual modes will suffer.
Special thanks
Thanks to ToonLucas22, who used the idea of a Tetris platform in his run. Thanks also to SmashManiac for encouraging me to optimize 1 piece further.
Baxter: Submission file replaced by a 10 frame faster version.
Baxter: A technically good movie, with a good audience response. It only completes one mode however, while there are several interesting modes to show off. A prefered version that completes several modes is already being worked on. This movie will be rejected in favor of the upcoming superior version.
