Astral Realms

I have run into major issues with the player collision. Collision between items and the spherical terrain works, but I cannot get player collision working. The collisions mesh is visible, however the player passes straight through. I am in the process of debugging this, but it is taking more time than I thought.

I believe the problem is that the player code is excluding the spherical terrain object from consideration during the collision checks. The collision code in both item and player is however, complex, and so I am having issues tracking down where this is happening. I am, nonetheless,hoping to get player collision working in the next few days.

The next goal is to add paging and scalable terrains, and test the gravity and 3D player rotation code which I implemented some time ago.

Then I can begin the next phase of the Virtual World project, which I will announce once the spherical terrain is working the way I want it to.

I have finally made some progress with the collision system, after having been stuck for some time. I have managed to get a collision mesh created, and item collision working, but player collision is still not working. I am very close to figuring this out.

I tried various different ideas to get collision working, from modifying the existing terrain collision system, to copying the polysoup/OPCODE solution from the tsstatic class, and anything and everything in between.

Eventually, I realised that I needed a collision class that handled just a raw triangle, ie, three points. The existing collision types were for handling more complex shapes, such as meshes, terrains, or boxes.

I eventually found some code to allow me to implement such as system on a T3D archive site HERE. This is actually a resource to implement a road object in T3D, but it contains a class called “TriangleConvex” which is a collision type designed to take only a triangle object. I was able to easily implement and apply this to my terrain.

Now, I just need to fix player collision, and then implement optimisation. As can be seen from the images below, I have significantly reduced the complexity of the terrain, since every vertex is now being added to the collision list. I need to only add the vertices close to the player to the collision object, and update these vertices as the player moves. This will allow for high resolution terrain which also runs fast.

I believe I have fixed the seaming problem and generated a sphere displaced with seamless noise.

I was looking at this problem the wrong way. I was looking at the sphere as a combination of six flat planes, and trying to apply 2D noise to each plane, similiar to the way a standard terrain would be created. This created seams between the six sides of the quad sphere.

However, even though the sphere is create from a projected cube, this is not the correct way to apply noise to it. I should have looked at it as a regular sphere, and used a 3D noise function to take in the x,y, and z coordinates of the sphere as inputs. This is what I have now done.

I am using libnoise, and opensource noise library. After generating the noise value using the coordinates of the sphere as inputs, I am then displacing each point on the sphere along it’s normal. This is the critical step that I was missing for a while. In order for the sphere points to displace in the correct direction, the normal needs to be taken into account.

The sphere looks more like an asteroid now, rather than a planet, but fixing that can be done by configuring the noise function.

The next step is to work on procedural texturing of the sphere. I intend to use the height value of the individual points to paint different textures, mountains would be while, low areas blue, areas in the middle green, etc etc.

I have made significant progress on the spherical terrain. I now have a noise function applied to the terrain, and I can generate a decent looking terrain surface.

This has led me to what I suspect will be the most difficult part of this project, which is aligning the seams of the planes. I need to find a way to join the seams together smoothly, without any visible distortion.

I am working on a solution involving linear interpolation. I intend to isolate groups of vertices near the edges of the planes, and determine which two seams are in contact with each other at each boundary. I can then get the heights of the vertices closest to the boundaries on both sides, and use linear interpolation to move the vertices in the boundary region in such a way as to smoothly join the two edges together.

Isolating the vertices has proven difficult, and I am still having some problems with that. I am now attempting to gain access to the data that I need in order to perform the interpolation. This required some reworking of my code, since I was using a single vertex list for each plane, filling it, then emptying it and using it again for the next plane. However, for the linear interpolation to work, I need to gain access to the vertex lists of multiple planes at once, so I needed to re-code the system to use a different vertex list for each plane.

This now works, and I am almost ready to test my linear interpolation concept.

I have completed the Quad Sphere! It took some time to draw the other five planes of the sphere in the correct location and the correct orientation, but that is now working.

The system is scalable, I can enter a resolution and it will draw a quad sphere with that resolution, so for example a resolution of 3 will produce a very blocky sphere, such as this:

while a resolution of 10 will produce a much higher resolution sphere, like this:

The resolution gives the number of vertices per plane, so a resolution of 3 works out to: 3*3 = 9 vertices per plane. In this grid, there are (3-1)(3-1) squares, which is 4 squares, each square contains two triangles, which is 8. There are 6 sides to the quad sphere, which means a final poly count of 8X6, which is 48. Clearly, the complexity scales dramatically with resolution, and this is why, for generating even reasonably large or complex terrains, optimisation and culling  will be extremely important.

The algorithm seems to be working well, although I am noticing some distortion of some kind visible around the seams. I am hoping that this is a cosmetic issue that can be fixed when I implement texturing, but it may be more serious. The next step is to apply a noise function to the vertices to produce an actual bona fide “terrain”. What I have now is really just a sphere. Once I do this, I will know how serious the seaming issue is.

I have succeeded in drawing one face of the quad sphere! I had made a slight error in the implementation of the mathematical equations, which is why it didn’t work before, but it is working now. This is a major step forward. I now need to draw all six faces of the cube, so that I can project them to form the completed quad sphere. In the images below I have copied and pasted several several faces and manually rotated them, to confirm that the edges line up correctly.

After I get the complete sphere drawn, I will need to apply a noise function to produce the actual detail of the terrain. With T3D’s terrain, this was easy, but with my new solution I am doing this completely from scratch, so I will likely need to apply some kind of smoothing function manually.

Then, I will need to look into texturing, collision detection, culling, and a whole host of optimisations!

After the issues I had with modifying T3D’s terrain, I had to come up with another solution to the problem of creating a quad cube. In the engine there is a piece of sample code called the “RenderMeshExample”. This basically outlines the code necessary to create and render a mesh using vertex and primitive buffers. I have been modifying this code in order to create a six sided cube with a specified number of vertices on each side.

I then intend to apply the sphere mapping equations to the cube’s verts, to produce the quad cube. Using rendermeshexample as a starting point has one main advantage, in addition to allowing me to write the rendering code from scratch. It allows me to create all six sides of the cube within the same mesh, so that there should be no visible seaming effect. This avoids a very difficult problem that I would have had with my previous solution.

However, I will have to write collision, optimisation, culling, paging, and a whole host of other features myself, since rendermeshexample doesn’t contain them. This will certainly prove difficult.

I have managed to draw an entire face of the cube, and change the resolution, creating a grid of whatever size I want. I now need to project this face to a sphere, and of course, draw the other faces.