SRL Archive - MSI Bugs

Alright so it's only a matter of time before every tree in RS is updated, so we really need to figure out a better way to find and click these trees. Since only the trunk is "click-able", we're pretty much SOL for Oak, Yew, and some normal trees (don't know what Maples/Magic look like with new graphics, but I can imagine they'll be just like Oaks and Yews).

With the new TLocation implementation we will be able to adjust the camera angle to different angles to find and click the trunk; however, this may turn out to be a lot slower than we want. You could argue that this method would be more human-like, though, which is definitely a plus.

Anyway, we open to any kind of suggestions about how to do this, no matter how stupid you may think it is. Here's a list of what's been suggested/what I've thought about:

• Making MMToMS 100% accurate and use SPS points to find trees through the minimap. This would mean SPS would also need to be 100% accurate.

Post the first thing that pops into you head. ;)

Thanks,
The MSI Team

P.S. If you say to use Reflection I'll rip your head off and change your rep to -9001. :rolleyes:

One thing: Use reflection for other things than MSI!

But seriously, making SPS 100% accurate would include SPS being much slower than it is today and I've never really liked MMToMS as a way of finding objects :/

The first thing that popped into my head was this:
1. Find trunk color.
2. Find tree color.
3. Compare these two TPA's to eachother, figuring out the most likely place for a tree.

This too would be a little slow, but I really don't know a good solution to this

EDIT: To whoever de-repped me for this post: learn to read white text… If it was you Coh3n then I'm kinda disappointed you actually believed me :(Was only worth 1 rep point so it couldn't be Coh3n

1. Tile of trunk
2. Search for trunk color in that tile
3. Click

Uses reflection but will be accurate.

From: Zyt3x
One thing: Use reflection for other things than MSI!

But seriously, making SPS 100% accurate would include SPS being much slower than it is today and I've never really liked MMToMS as a way of finding objects :/

The first thing that popped into my head was this:
1. Find trunk color.
2. Find tree color.
3. Compare these two TPA's to eachother, figuring out the most likely place for a tree.

This too would be a little slow, but I really don't know a good solution to this

EDIT: To whoever de-repped me for this post: learn to read white text… If it was you Coh3n then I'm kinda disappointed you actually believed me :(Was only worth 1 rep point so it couldn't be Coh3n

I think Nava said something like that a while ago when we were talking about this. I don't think it would work well for Oaks and Yews, though as you can't really see the trunk as the leaves are pretty thick, and if we changed the camera angle, the leaves wouldn't be found.

From: Dynamite
1. Tile of trunk
2. Search for trunk color in that tile
3. Click

Uses reflection but will be accurate.

No offense, but that's completely pointless (don't mean to sound harsh). If you're going to use reflection for something like that, you may as well just use models or objects.

From: Coh3n
I think Nava said something like that a while ago when we were talking about this. I don't think it would work well for Oaks and Yews, though as you can't really see the trunk as the leaves are pretty thick, and if we changed the camera angle, the leaves wouldn't be found.

We could develop some new high-tech method of finding objects using edge-finding and whatnot instead of the normal color find method. After all, human vision is based on object outline rather than content (Example: You can sltil raed tihs plterecfy fnie bcuasee yuor viison is bsaed on the "form" of the ojbcet, raethr tahn the atucal lteter aanrgeemrnt)

From: Zyt3x
We could develop some new high-tech method of finding objects using edge-finding and whatnot instead of the normal color find method. After all, human vision is based on object outline rather than content (Example: You can sltil raed tihs plterecfy fnie bcuasee yuor viison is bsaed on the "form" of the ojbcet, raethr tahn the atucal lteter aanrgeemrnt)

Woooow. :p

Neat concept, but how can it be applied to objects?

From: Coh3n
Woooow. :p

Neat concept, but how can it be applied to objects?

Let's see…
http://en.wikipedia.org/wiki/Edge_detection
http://en.wikipedia.org/wiki/Canny_edge_detector

Then we make it search for a shape 'like' a tree

Did a little testing with an online edge detector (http://matlabserver.cs.rug.nl/cannyedgedetectionweb/web/index.html).

http://i.imgur.com/cZOIt.png

^ Original

http://i.imgur.com/RTMlG.png

^ Canny'd

The most difficult part is understand what is a tree and what's not. Changing angles will (obviously) drastically change the look.

The little 'trick' that Zyt3x pointed out isn't actually true, but I'll still compare it to that for familiarity.

The words are very clearly defined and have a very small margin of change. They can only change so many ways (which increases, obviously, as the characters in a word increases), which is why we can understand them. Our brains work quite fast and can figure it out.

However, with something like trees and edge detection, there are far too many ways that the edges can change to be effectively found by a program. The variation is in three dimensions, so rather than a simple change like 'would' to 'wluod,' which is easy to figure out for a program/human, you'd have to add another dimension (and take off the first/last character requirement), which would give you something like 'dewlfopu'. All of the characters are there for the word 'would,' but there's extra information that is nearly impossible to account for and determine what is important and what isn't.

And in reality, it's much more difficult than only three added characters of information.

So, not to be a downer, but Zyt3x, that's not a good route. And no, humans do not understand things based on the outline and not the content.

Looks like we'll use something else than http://matlabserver.cs.rug.nl/cannyedgedetectionweb/web/index.html then

This was actually taken care of in the TLocaion update. :) Works pretty well in my opinion.

I didn't wanna make a new thread so thought I would post here.
I started using MSI finally :p haha. While using it at east varrock oaks it just stood at the tree for like 2-3 mins and didn't do anything. It did work for a few loads tho.
As I got to go out soon, I can't test again but i will late when I'm back, to double check.
In the mean time maybe someone else could try it out to see if they have the same problem.

P.S When doing home tele you should make it walk to the musicians on the way to where ever its going. Just a thought, as it will speed things up slightly :p

From: Jakkle
I didn't wanna make a new thread so thought I would post here.
I started using MSI finally :p haha. While using it at east varrock oaks it just stood at the tree for like 2-3 mins and didn't do anything. It did work for a few loads tho.
As I got to go out soon, I can't test again but i will late when I'm back, to double check.
In the mean time maybe someone else could try it out to see if they have the same problem.

P.S When doing home tele you should make it walk to the musicians on the way to where ever its going. Just a thought, as it will speed things up slightly :p

That's one of the locations I actually have wrote down as broken, perhaps that was why. Regardless, I can probably fix it this afternoon.

I'm guessing the object record just needs to be edited. Oh and NCDS, load the .aca file in Dev Tools/ACA so the current colours are still included.

Although this topic is resolved, I seem to still have an issue with "tree finding" My simba cannot seem to differentiate between a regular tree/oak/willow.

What I propose
Create model images of the trees, for example Willow tree Take EVERY type of willow and create a multi-model pack named Willow. (I would suggest a 3d image of it, it takes more time but this way the bot doesn't constantly need to be in one point of view)

Call that pack every time you want to cut willows, the bot will ignore other trees if they do not follow the same model size/color/shape

I see this working better than JUST color because runescape tries really hard to blend colors in multiple trees to make it hard for color only bots to operate.

(mind you I have yet to look at any tutorials for coding etc so my logic may not be a possible outlook. )

-R4

From: R4nd0m
Although this topic is resolved, I seem to still have an issue with "tree finding" My simba cannot seem to differentiate between a regular tree/oak/willow.

What I propose
Create model images of the trees, for example Willow tree Take EVERY type of willow and create a multi-model pack named Willow. (I would suggest a 3d image of it, it takes more time but this way the bot doesn't constantly need to be in one point of view)

Call that pack every time you want to cut willows, the bot will ignore other trees if they do not follow the same model size/color/shape

I see this working better than JUST color because runescape tries really hard to blend colors in multiple trees to make it hard for color only bots to operate.

(mind you I have yet to look at any tutorials for coding etc so my logic may not be a possible outlook. )

-R4

Or an easier way:
findtree->Hover tree->search uptext for that certain tree->act accordingly.

From: R4nd0m
Although this topic is resolved, I seem to still have an issue with "tree finding" My simba cannot seem to differentiate between a regular tree/oak/willow.

What I propose
Create model images of the trees, for example Willow tree Take EVERY type of willow and create a multi-model pack named Willow. (I would suggest a 3d image of it, it takes more time but this way the bot doesn't constantly need to be in one point of view)

Call that pack every time you want to cut willows, the bot will ignore other trees if they do not follow the same model size/color/shape

I see this working better than JUST color because runescape tries really hard to blend colors in multiple trees to make it hard for color only bots to operate.

(mind you I have yet to look at any tutorials for coding etc so my logic may not be a possible outlook. )

-R4

I like the idea, I just have NO idea how to go about doing it, or if it's even possible. But you're right, we do need a more accurate tree finding method. Oaks and Yews are the worst because the leaves are so huge and not all parts of the leaves are "clickable" if you know what I mean.

Willows should work just fine, and same with dead trees. Normals I'm not 100% sure about. Likewise with Magics and Maples.

From: kingarabian
Or an easier way:
findtree->Hover tree->search uptext for that certain tree->act accordingly.

That's what we do now, and I'll admit MSI's tree finding isn't amazing since the trees were updated.