Physics engine – collisions, wrap up.
New version, with ∞ % more physics.
It took a few days but it’s here: the new physics engine that handles collision detection and collision response.
A screen shot.
I should try to make videos, they would show the physics better.
Figure 1 shows a screen shot from the last version. The tiles with a ‘3D’ border are raised: they are obstacles on which you collide and bounce more or less according on their material. The flat tiles have a friction coefficient that also depends on their material, which slows you down as you walk. Materials pictured here are sand, rock, shallow water, deep water and rubber (rubber is stupid but I wanted to try 100% bounce).
You may have noticed that the entities looked circular now: I represent their bounding box. I chose to give them a diameter of exactly one meter, which corresponds to the width of a tile. That allowed me to make sure that entities could squeeze in exactly in the tightest conditions.
Everything worked pretty much immediately which really makes me appreciate having spent some time doing the analysis in my previous posts.
One thing went wrong though: I implemented the position correction before implementing anything about velocity correction. Which is okay. I wanted to have a perfectly working position correction before moving on. I unfortunately included in the code a couple of lines that made perfect sense for a game with position correction only, but which bugged weirdly later with the velocity correction added: entities would stick to walls, but only sometimes, and they couldn’t fit in a one-meter-wide corridor, but sometimes they teleported to the end of that corridor. Creepy! It took me about one hour to find what was wrong :(. Once I removed the few lines, it all worked smoothly. I shall try to pay more attention to these kind of things: I should not polish a my code before it’s finished.
One step away from the design.
The implementation followed the design very closely. There is one difference though, and it has to do with a suggestion I made in my previous post in the section about Simultaneous collisions. I suggested that I could apply all the position corrections at the same time by adding their penetration vectors.
It wouldn’t have worked: I should not add everything. I should pick the biggest correction in X and the biggest in Y. It would force me to start projecting on some axis, and projections are expensive. Actually, they are more expensive than my collision detections (I’m lucky to have such a simple geometry). Also, I don’t even know what to do in case of corrections that cancel each other, because how do I define the biggest then ?
Do not worry if the previous paragraph confuses you. I hardly understand it myself: it’s messy, complicated, it treats position and velocity differently, and wouldn’t even make the code faster. So I ditched that. The code is so simple now!
Download the demo.
Have it your way:
- Download the zip at https://github.com/Niriel/Infiniworld/zipball/v0.0.5
- or clone/pull/checkout the last tag with
git clone firstname.lastname@example.org:Niriel/Infiniworld.git git pull git checkout v0.0.5
- I dislike the tangential efficiency applied in collisions. You can feel it when you move in diagonal against a wall: a kind of friction is there, slowing you down. It’s particularly strong on waterfalls (raised water tiles) but you can feel it on piles of sand or other tiles too. Although it makes sense to be slowed down when you run against a wall, it’s just annoying to play. I think I will just set this tangential efficiency to 1.0 unless I really need it different.
- I do not make any difference between entities that walk and entities that fly or entities that are pushed (like a boulder). The friction to apply is very different in these cases. The engine does not handle it at all.
- The friction of the floor slows you down, but there is nothing that makes you have difficulties starting. Imagine sanding on ice: the low friction makes it hard to stop –which my engine can reproduce– but the same low friction reduces the grip of your soles and steals some of your walking force –which my engine does not handle at all–.
I will implement the point 1 because it’s just a matter of changing a few constants. However I shall not on the points 2 and 3 in the near future. I think that we have enough physics to make things interesting: we can move around and push things. Now that we can explore, we need something to explore. It is time to spend some time on procedural content generation \o/.