anyweez / glimpse Goto Github PK

Similar to problem described here: https://azgaar.wordpress.com/2017/04/03/coastline/

Interpolation on coastlines looks pretty good and I want to keep it, but the fill is still being performed on pre-interpolation polygons which leads to some visual artifacts.

The post above describes one way to fix this with svg masks, but my first attempt will be to try to render landforms as single complex polygons. I expect the challenge will be rendering lakes but I think this will be a simpler solution overall if I can make it work.

Humans should have an easier time surviving in certain biomes.

@anyweez to elaborate

The forest plugin seems to be adding forests to oceans, sometimes multiple times. Likely cause is we're not filtering out water cells as starting positions for forest generation.

Support more realistic terrain generation for larger worlds, as well as for identifying realistic locations for forests (#13) and potentially future formations like deserts.

Initial implementation will likely use latitude (assumes the world has a consistent equator) and elevation to determine a temperature distribution for a cell.

The language server is responsible for providing names for entities in any gen.py instances. Since each language has to have a training period, the idea is that this will allow the language server to be long-running and avoid training periods.

Once we've got towns (#15), we should add roads between them.

At this point, I'm thinking properties of towns will be derived based on the desirability / survivability of a place and that roads will inherit that, vs roads having any direct impact on the towns themselves. I think it should still "look right" in the simulation (bigger cities will have more roads, but not because the roads cause the growth of the city).

I'm seeing some lower quality (unpronounceable in english) names come through, as well as some that clearly lack context of where the city is ("Gully Harbor" isn't anywhere near water).

Since the latest refactor, tectonic plates no longer affect cell elevation. Simplex noise is the primary contributor to elevation, which lacks the "continental" look of the previous plate-based approach.

For reference, here are the Earth's tectonic plates:
https://en.wikipedia.org/wiki/Plate_tectonics#/media/File:Plates_tect2_en.svg

There are two specific properties of plate tectonics that I want to try out:

1. Individual plates may be higher in elevation than others. All cells of certain plates should start with a base elevation.
2. Each plate boundary may have certain elevation-impacting features, depending on the boundary type (divergent, convergent, transform).

The old site linked from github (https://homeworld.surge.sh) is no longer relevant, and I'd like for there to be a project page of some sort.

Goals:

• Describe what Glimpse does
• Provide a gallery of examples, including from older versions
• Capture lightweight release notes for major/minor versions

Right now rivers are not being rendered at all. Step 1 = make sure they're being calculated correctly, Step 2 = improve render rendering from previous iterations.

Forests should be biome-aware. There are two types of forest biomes, which should have denser tree growth.

Look into Poisson disk sampling for this use case:
http://devmag.org.za/2009/05/03/poisson-disk-sampling/

60fps rendering is already causing a lot of extra work without any sort of population spawning (#2). If things get any more complicated we'll need to parallelize; it's honestly probably a good idea anyway once we start migrating to have a more complex UI.

World, Cell, and Population should all be pretty testable. There are also a lot of crazy ass dynamics that I'd like to have a constant eye on.

Basic components exist as of b8e4776 but I need to do some testing in the core application. Right now they're independent.

Current plan for population spawning is pretty rudimentary but I think it'll be fine over time w/ law of large numbers.

1. Periodically spawn a random population and a random location. Do this pretty regularly (every 1 - 10 seconds).
2. Periodically evolve populations with they migrate. Need to change names of the population when this occurs.
3. Likely need to look into web workers to handle some of this work (#3).

Need to figure this out for rendering the printed version.

One possibility:
http://devmag.org.za/2009/05/03/poisson-disk-sampling/

Right now I'm just using a random distribution by biome type which looks pretty dull, especially when most of the map is forest.

Need to figure out how to render the depth of lakes -- oceans use WaterlineHeight, but this doesn't work for lakes which form at elevations above sea level.

I think each lake may need its own calculated waterline height, and depth can be determined relative to that.

Put together one command to generate and render X worlds. Performance doesn't matter at this point.

We need a quantitive way to assess the quality of glimpse worlds to verify whether a set of changes are net positive or negative. This may change over time.

Initial proposal

Train a system that can provide a quality score for a given map. This score is likely to be more about aesthetics than anything else, so it may resemble an image classification problem (example to follow). The general implementation of testing the quality of a particular branch would be:

1. One-time: manually label output maps based on their quality (high/low or a continuous score 0-1, most likely). Build up a training set of these examples -- they made need to be updated when major new features are added.
2. One-time: train a model to score/classify new worldmaps.
3. To test a set of changes, generate many (1000?) worldmaps and use the model from 2 to evaluate them. Its important to observe both the % of high quality items as well as average quality, with the former likely being more important since we can filter out low-quality maps using the same score.

Need to do a quality pass on forest, mountain, and hill iconography. Supersedes #37.

Distributions are decent and may not need adjustments but the icons themselves aren't good enough yet.

1. Ideally there's some minor variance between instances (each tree doesn't look exactly the same).
2. Better designs overall. In the case of mountains there's nothing at the moment.

This code is very functional w/ pretty explicit types and I'd benefit from some type checking.

New worlds use a variety of important variables which are not currently exposed to users. These should be configurable.

Examples:

• Aquifer depth
• World size
• Altitude variance

There will likely be others in the future. Note: at this point there's no user-exposed "seed" that will allow you to generate the same map deterministically, and its unclear that we'll ever include this.

Following up on #32 there are still some improvements I can make to improve rendering quality of rivers.

1. Each river segment (cell edge) is still a straight line. Add in some meandering to make it look more natural.
2. Each segment's width is determined independently, and there can be sharp increases in width between segments.
3. Due to interpolation with coastline rendering, rivers don't always visually terminate on coastlines (though the data structure suggests they will).

Useful notes:
https://azgaar.wordpress.com/2017/05/27/confluences/
https://azgaar.wordpress.com/2017/05/08/river-systems/

Either Angular 2, React, or Ampersand.

Any named entity should have it's name rendered. Optional but nice touch would be to approximate city size so bigger cities can have bigger text.

Currently these two can run in either order, but this means that its possible to have a forest-filled lake. This is happening because the lake plugin is modifying the celltype cellprops, which it really shouldn't be doing.

Need to figure out how to resolve this, though I think the best solution will be to merge lake gen into the terrain plugin where cp_celltype is being set already.

Deep water vs water, snow-capped mountains vs standard mountains, and a few different colors of grass.

Forests and/or hilly areas can look like a mess when icons start overlapping, and I think some small visual tweaks like removing partial transparency will help

https://www.reddit.com/r/proceduralgeneration/comments/7natln/resources_or_methods_for_biome_generation/

Identify areas that are desirable for humans (or potentially another species). A subset of these locations should have towns, and towns should have names.