Automated generation of Manic Miner speedrun/walkthrough

[jetsetdanny]

On 4/13/2021 at 11:13 PM, jetsetdanny said:

The page features a full screenshot gallery, but if Norman Sword follows Ian's suggestion expressed above, I will be happy to update the screenshots accordingly.

 

  

11 hours ago, Norman Sword said:

I took up the challenge. Apart from the full size main screen pictures, I have changed every single screen.

The auto play in only one cavern is different.  (Amoebatrons' Revenge.)
LASTv2.tap

 

My statement was a promise, not a challenge, made in the context of Ian's suggestion of inserting the alternate sprites into 'Amoebatrons' Revenge' and different Fire cell/item bitmaps in 'Processing Plant' - i.e. changes in two rooms, not all of them.

Still, I will keep my promise, but not immediately, but after a cooling-off period when I decide that a release of another version is not likely. Spending my time taking screenshots only to have to re-take them a couple of days later is not my idea of proper use of JSW-time.

Unless you declare solemnly that you have no specific plans to release another upgrade - in which case I might force myself to do the screenshot-taking chore for the whole game again 😉.

[jetsetdanny]

6 hours ago, MtM said:

I don't know if this would improve it Danny, and I am not trying to create work, but maybe a table in the form

of something like this

 

            Best human score - Best cpu score

Central Cavern  20000     -  21000

Cold Store         20000     -  21000

 

You get the idea. No worries if you think it no use etc, it isn't really, just interested personally to see how far behind human players are in terms of best scores etc. I guess there is something to be learned by speedrunners from that maybe, or just people wanting to improve their playing.

 

Thanks for this idea, Mtm 🙂 .

No, I won't implement it, for several reasons:

- It wouldn't belong to the "Manic Miner: creme de la creme" page, because a human can't play the regular rooms in "Manic Miner: creme de la creme". So it's not possible to compare human and CPU performance there.

- Such a more general comparison might be perhaps more appropriate on the Manic Miner page, page I don't want to go into this kind of detail there.

- There is no difference between the best (Rollback-assisted) human performance and the CPU's performance in 18 of the 20 caverns of the original MM 😁 . There is still a 5-point difference in "The Warehouse" and a 12-point difference in "Solar Power Generator", that's all. See here for more info or here for the download which documents best human (Rollback-assisted) performance. So there would be little to compare in the table you suggest 🙂 .

[Norman Sword]

The Bug Byte pictures. 

 

Edited April 16, 2021 by Norman Sword
Removing one picture

[Norman Sword]

Software Project screens.

Edited April 16, 2021 by Norman Sword
adding one picture

[jetsetdanny]

Thank you, Norman Sword! 👍 I appreciate your contribution.

I will use most of these screenshots and update JSW Central soon.

I might re-take screenshots of a few caverns for the following reason:

I generally tend to take screenshots soon after entering a cavern, so that it can be seen in "all its glory", i.e. without the crumbly cells crumbled, without any items collected yet, etc.

Due to this, in "creme de la creme" the visual representation of Willy's movements on the status bar is not displayed fully - just the beginning of it.

There are some caverns (like the Amoebatron ones) where even after playing for quite a bit no "glory" is lost, while the whole status bar is filled with the visually represented movements.

I will retake screenshots of these caverns so that the visual representation of Willy's movement is seen all along the bottom of the screen at least in some screenshots.

[IRF]

2 hours ago, jetsetdanny said:

- There is no difference between the best (Rollback-assisted) human performance and the CPU's performance in 18 of the 20 caverns of the original MM 😁 . There is still a 5-point difference in "The Warehouse" and a 12-point difference in "Solar Power Generator"

19 of the 20 caverns, actually - as I reported earlier in this thread, I managed to match crem's algorithm's performance in both the BB version of 'The Warehouse' (via two different routes!), and the SP version of that cavern.  Although I didn't create any recordings of myself doing so (I did it in the QAOP online emulator, which allows a single rollback), and I haven't (yet) updated the 'Highscore Challenge' table to reflect my achievement in this regard.

I haven't managed to beat either the algorithm's or your score in 'Solar Power Generator' though.

[jetsetdanny]

Well done, Ian! 👍

Sorry, I missed that information earlier. So there's even less reason to create a comparison table between humans and CPUs 😊.

[IRF]

I reported my success with the 'alternative' route through the Warehouse in this post (which you 'liked' at the time!):

Then in the post which follows immediately after that, I discussed how the algorithm had taught me the most efficient way to proceed diagonally down through a crumbly field - although having looked again at that post, I can see that I didn't actually explicitly state that during the course of my investigations, I managed to match the algorithm's best Warehouse score via that route (though I did do so!)

http://jswmm.co.uk/topic/580-automated-generation-of-manic-miner-speedrunwalkthrough/?do=findComment&comment=13141

Edited April 17, 2021 by IRF

[IRF]

I also created some screenshots at the time of my Warehouse investigations, but I don't think I've posted them before now.  Here are those screenshots, with some further analysis (if you can stay awake long enough to read it!)

The screenshots are named after the final scores accrued in each attempt - in the order that I have uploaded them here, namely: 'Warehouse 1948', 'Warehouse 1949' and 'Warehouse 1951 - Best Final Score'.

All routes are identical, apart from the sequence of movements in the central crumbly field to the right of the yellow guardian.

The significant differences in keypresses took place just after Willy jumped rightwards whilst he was underneath the yellow guardian (collecting the item above the conveyor in mid-jump).  In all cases, upon landing after that jump, Willy ended up above the 'ar' of "Warehouse" in the cavern name, in his right-facing animation-frame 3 (N.B. the Right movement button was depressed upon landing so that the left-right movement flag stayed set; otherwise he would have settled in right-facing frame 2).

- For the score of 1948, upon landing after that jump, Willy walked rightwards by 8 frames until he was in his right-facing animation-frame 3 above the 'eh' of "Warehouse" - just about to cross the cell boundary that would take him above the 'ho' of "Warehouse" - then waited until he dropped down through two rows of crumblies (this took 19 frames in total - 5 to clear remaining pixel-rows in the first row of crumblies, then 3 frames for Willy to drop through to the next cell-row down - he 'hovers' in thin air for a single frame before starting to descend by half a cell-row per frame - then another 8 frames for the 8 pixel-rows of the next pair of crumblies, then 3 more frames for Willy to drop down another cell-row), and then upon landing he immediately jumped rightwards.  27 frames elapsed between landing after the jump in bold and taking the next jump.

- For the score of 1949, upon landing after the jump referred to in bold, Willy came to a halt in right-facing animation-frame 3 (N.B. I had to get the timing of letting go of the Right button exactly spot on, so that he carried on moving at the end of the jump but he did NOT proceed across the boundary into the next cell-column), then he stayed still in the pair of cell-columns above the 'ar' of "Warehouse", waiting for 10 more frames (7 to crumble the remaining pixel-rows and 3 for Willy to drop down a cell-row) until he dropped down through one row of crumblies, then walked right by 4 frames (taking him above 're') and waited there for 8 frames (5 to crumble the remaining pixel-rows and 3 to drop down by a cell-row) to drop down through another row of crumblies, then he walked rightwards by another 4 frames before jumping rightwards.  (In those last four frames before the jump, the first frame was spent crossing the cell boundary from above 're' to above 'eh', with the remaining three frames spent above 'eh' - that's why the crumbly above 'e' has one more pixel-row crumbled away than the crumbly above 'h'.)  26 frames elapsed between landing after the jump in bold and taking the next jump.

- For the score of 1951, upon landing after the jump referred to above in bold, Willy walked forward by 4 frames (so he was in his right-facing animation-frame 3, above the 're' of "Warehouse"), waited 8 frames to drop down through a row of crumblies, then walked forward by 4 frames (into right-facing animation-frame 3, above 'ho'), waited for another 8 frames to drop down through another row of crumblies, and then upon landing he immediately jumped rightwards.  24 frames elapsed between landing after the jump in bold and taking the next jump.

[After that latter jump, which took Willy underneath the white guardian, the route is once again identical in all three cases.]

P.S. One more point to make: In all three screenshots, eight crumbly blocks have been trampled to some degree within that central crumbly field (and in all three cases, four of those blocks have completely crumbled to air), but in the most efficient route (which led to the final score of 1951 points), Willy had minimal contact with three of those eight crumbly blocks (i.e. three crumblies have only lost a single pixel-row).  That is the key to obtaining the fastest route across/down through a field of crumbly blocks - ensure that each time you drop down by a cell-row, Willy is in position to immediately start walking forwards into the next cell-column, so that the amount of time spent idly waiting to drop down to the next row is minimised.

 

 

Edited April 18, 2021 by IRF

[jetsetdanny]

Thank you for your detailed explanations and screenshots, Ian! Congratulations on both achieving these scores and explaining them so precisely! 👍

I can't pretend I remember every single post I've 'liked' (with over 13,000 posts published on the forum so far; I don't know how many I've 'liked', but I would think it may be every third one perhaps, so over 4,000 😄).

For the record, there may be some posts that I've 'liked' that I don't even understand. There are, e.g., detailed technical explanations offered here and there that I truly like - I believe it's impressive someone was able to come up with them and took the time and trouble to share them, but that doesn't mean that I'm able to understand their contents fully! 😮