Improve pathfinding

[suprohub]

This pr adds advanced parkour system, where bot can jump in all directions

[mat-1]

Thanks!

Though, I'd recommend splitting up the changes that you're planning into more PRs to make for cleaner diffs and to make them easier for me to review.

Also, fyi, I've been holding off on making large updates to the pathfinder until there's a realistic benchmark for it. Azalea's pathfinder is several times faster than Baritone in some cases and I'd kind of like to keep it that way. :3

This is why pathfinding through water isn't implemented yet -- it's fairly easy to implement with custom moves but I think doing it without hurting performance might be hard, and I'd like to have an easy way to check that.

There are a couple benchmarks for the pathfinder, and they are useful for checking some types of changes, but they're not realistic. I think a good way to add a realistic benchmark would be by copying a region file (or a few) from a Minecraft world into Azalea's benches directory, and having a benchmark that parses that region file into Azalea's in-memory world representation and calculates a path over it.

[suprohub]

Ok

[suprohub]

I think litematic is more good becas you can use only special regions and not import all anvil regions which is ver big 💀
I tested, maybe works.

noviui@suprohub ~/R/azalea (improved-pathfinder) [SIGINT]> cargo bench --bench realistic_one
    Finished `bench` profile [optimized + debuginfo] target(s) in 0.16s
     Running benches/realistic_one.rs (target/release/deps/realistic_one-4063cdf73c0993a9)
Bench: Getting over it, Start: 103 6 120, End: 11 86 15
Benchmarking Getting over it: Warming up for 3.0000 s
Warning: Unable to complete 100 samples in 5.0s. You may wish to increase target time to 8.1s, or reduce sample count to 60.
Getting over it         time:   [80.290 ms 80.502 ms 80.738 ms]
                        change: [−6.6241% −4.4933% −2.9198%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 1 outliers among 100 measurements (1.00%)
  1 (1.00%) high severe

Bench: Tiny, Start: 1 1 0, End: 1 1 23
Tiny                    time:   [2.2922 ms 2.2978 ms 2.3050 ms]
                        change: [−13.021% −8.9040% −5.6421%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 13 outliers among 100 measurements (13.00%)
  1 (1.00%) low mild
  3 (3.00%) high mild
  9 (9.00%) high severe

Bench: The Jumbo LABYRINTH (by TriscuitsAllDay), Start: 105 1 109, End: 2 1 46
The Jumbo LABYRINTH (by TriscuitsAllDay)
                        time:   [47.741 ms 47.847 ms 47.975 ms]
Found 10 outliers among 100 measurements (10.00%)
  6 (6.00%) high mild
  4 (4.00%) high severe

idk how it knows what performance was improved maybe i randomly run 2 times and forgot

[suprohub]

Tiny is tiny parkour map, getting over it its map where you need climb on big mountain and labyrinth its just labyrinth

[suprohub]

Will be cool make func what predicts capacity of call_successor_fn vec:

let mut edges = Vec::with_capacity(16);

Capacity 16 is no good. If i make 32 getting over it alredy go to 32% performance boost.
but 32 not good too for small pathfindings, so we need alghoritm what predicts approx complexety

[suprohub]

Where are you?

[mat-1]

Hi, I'm sorry, I'm currently pretty overwhelmed and do not have time to review this (also why I haven't made any substantial commits to Azalea in a few weeks). I'll review this when I have time.

[suprohub]

oka

[SolsticeSpectrum]

@suprohub while doing the PR you might wanna fix the parkour in general, because the current pathfinder can block the tick thread on complex terrain (sync patch/recalc), which causes rubberbanding/timeouts. Perhaps move it to AsyncComputeTaskPool?

[mat-1]

while doing the PR you might wanna fix the parkour in general, because the current pathfinder can block the tick thread on complex terrain (sync patch/recalc), which causes rubberbanding/timeouts. Perhaps move it to AsyncComputeTaskPool?

You might want to repost that comment as an issue and answer these questions there, but I'm curious what your setup is. Path patching is deliberately blocking (it has a very small timeout and is meant to be able to react instantly), and path recalculations are already done on a separate thread. Are you running in release mode (if not, is the problem still present if you do)? How many bots are you running? What hardware (CPU, mainly) are you using? What's the terrain that you're pathing over like?

[suprohub]

Hmm. After update, i see performance regressions.

Bench: Getting over it, Start: 103 6 120, End: 11 86 15
realistic benches/Getting over it
                        time:   [74.501 ms 74.742 ms 75.037 ms]
                        change: [+2.8993% +3.6307% +4.3258%] (p = 0.00 < 0.05)
                        Performance has regressed.
Found 4 outliers among 100 measurements (4.00%)
  2 (2.00%) high mild
  2 (2.00%) high severe
Bench: Tiny, Start: 1 1 0, End: 1 1 23
realistic benches/Tiny  time:   [2.1969 ms 2.1993 ms 2.2015 ms]
                        change: [+18.500% +19.034% +19.549%] (p = 0.00 < 0.05)
                        Performance has regressed.
Found 17 outliers among 100 measurements (17.00%)
  3 (3.00%) low severe
  14 (14.00%) low mild
Bench: The Jumbo LABYRINTH (by TriscuitsAllDay), Start: 105 1 109, End: 2 1 46
Benchmarking realistic benches/The Jumbo LABYRINTH (by TriscuitsAllDay): Collecting 100 samples in estimated 22.782 s (500 iteratio
realistic benches/The Jumbo LABYRINTH (by TriscuitsAllDay)
                        time:   [42.645 ms 42.695 ms 42.758 ms]
                        change: [+3.1994% +3.5568% +3.9222%] (p = 0.00 < 0.05)
                        Performance has regressed.
Found 7 outliers among 100 measurements (7.00%)
  4 (4.00%) high mild
  3 (3.00%) high severe

noviui@suprohub ~/R/azalea (improved-pathfinder)> 

[mat-1]

I appreciate your work and patience, but I'm closing for now since this change makes pathfinding significantly slower (approx half the speed) for not much benefit. I will note that bots can implement their own custom moves by using start_goto_with_opts if they desire, so the code here could still be used by custom bots.

Also, I've implemented a more realistic benchmarking setup in a separate repo (which will be published soon).