Terrain for Panda3D

Just throwing some academic papers at the GPU to see what sticks. The goal is to have a tool to generate and render beautiful natural landscapes.

Current state: Alpha

• The lighting model is Lambertian diffusion with a static sun vector, the colors code-defined. It's basically the simplest hacked-together renderer imaginable.
• The environment model is trivial.
  • There is terrain generated from Perlin noise.
  • There is flowing water with rain and a fountain as a source.
• The simulation code is not optimized one bit.

Installation

Requirements: pip install panda3d Jinja2

For this project itself... Clone the repo and addit to your virtualenv, I guess? I haven't packaged anything yet.

Usage

python main.py

There are several options, documented in python main.py -h.

The turbulence in frame time (globalClock.dt in Panda3D) introduces turbulence in the simulation, so if you want a simulation that converges on a mostly steady state, it is recommended to set the timestep to approximate your framerate (assuming that you are running at a resolution at which the simulation reaches realtime performance), e.g. python main.py -t 0.01666 for 60 frames per second.

Current Model

• Hypermodel
  • boundary_condition: Whether the edges of the simulation are...
    • OPEN: Water flows over the edge as if the terrain and water heights were 0.0 beyond it.
    • CLOSED: No water flows beyond the edge.
    • WRAPPING: Water flowing out of one edge of the map gets added on the opposite edge.
• Map Data
  • terrain_height: Height of the terrain
  • water_height: Current height of the water
  • water_influx: Water influx rate water column to simulate e.g. rain, or fountains. Values are the water column added per second.
• Intermediate Maps
  • water_height_after_influx
  • water_crossflux
    • r to -u
    • g to +u
    • b to -v
    • a to +v
  • water_velocity
    • r along u
    • g along v
  • water_height_after_crossflux
  • water_height_after_evaporation
  • terrain_normal_map: Surface normals in GLSL encoding.
  • water_normal_map: Surface normals in GLSL encoding.
• Scalar model parameters
  • cell_distance: Distance between centers of neighboring cells; Default 1.0
  • pipe_coefficient: gravity g * pipe cross area A / pipe length l; Default 98.1 (Earth gravity 9.81 m/s**2, pipe cross area 10 m**2)
  • evaporation_constant: Percentage of water that would evaporate in one second if the amount of evaporation was constant. 0.0 for no evaporation. Default 0.05
• Frame parameters
  • dt: time step. The amount of time for which the simulation will be advanced.
• Process
  • Influx of water: Add new water to the simulation. water_height_after_influx = water_height + water_influx * dt
  • Calculate outflux For each direction/channel: water_outflux = current_outflux + height_difference * hight_difference * pipe_coefficient * dt, lower border of 0.0. If the amount of outflowing water is greater than that of water actually present in this cell, scale the outflux so that exactly zero water remains.
  • Crossflux of water water_height_after_crossflux = water_height_after_influx - own water_crossflux + neighbors' water_crossflux to this cell
  • Evaporate water: Remove a fraction of water from each cell. water_height_after_evaporation = water_height_after_crossflux * (1 - evaporation_constant * dt)
  • Update main data waterHeight = waterHeightAfterEvaporation

TODO

Current hot topics

• Bug: Why is water leaking out of a CLOSED/WRAPPING simulation?
• Simulation model
  • Step: Lateral sediment transport
  • Boundary conditions: Erosion is pretty whack at boundaries.
  • Steps: Sum up total terrain / water (to analyze the above bug).

Small nice-to-haves

• Aesthetics
  • Specular highlights
  • Fake SSS based on water depth

Icebox

• Aesthetics
  • Water
    • Side walls on closed/wrapped boundary
    • Waterfall on open boundary
  • Terrain
    • Side walls
• Hyper parameters
  • Initial data
    • Zero (current)
    • Generator shaders
    • Loaded images
• Simulation model
  • More papers
• Performance
  • Measure, and find the maximum simulation size / optimal workgroup size for realtime.
    • Requires: Make workgroup size a hyper parameter
  • Optimizations
    • Combine shaders into one
    • Instead of using array, use vector/matrix.
    • set up shared array for the work group that is 2 elements larger than the workgroup, making a one-element border around it. Preload global data into it, and do math. Write output back into global.
    • Red/Black mode
• Loading/saving simulations / dumping images
• Offline rendering, sidestepping the vsync limit
• Tiled simulation for offline rendering of super-large worlds
• Inflow map: R=inflow measured in height of water column, G=inflow by volume
• Outflow clamp map: Finer control over where and how boundary conditions occur.

Papers

• "Fast Hydraulic Erosion Simulation and Visualization on GPU": https://xing-mei.github.io/files/erosion.pdf
  • Overview
    • Influx of water
    • Calculating the pressure-based outflux from each cell to its neighbors
    • Update water height
    • Calculate water velocity
    • Erode material from terrain into water, or deposit it
    • Transport sediment through water flow
    • Evaporate water
  • Implementation
    • DONE: Increase water
    • DONE: Compute outflux flow; We treat neighboring cells as if they were connected by pipes.
    • DONE: Update water surface
    • DONE: Update velocity field
    • DONE: Erosion-deposition
    • Transport sediment new suspended sediment amount = s1 at position - uv * dt, interpolating the four nearest neighbors
    • DONE: Evaporate water
• "Fast Hydraulic and Thermal Erosion on GPU": http://diglib.eg.org/bitstream/handle/10.2312/EG2011.short.057-060/057-060.pdf?sequence=1
• "Hydraulic Erosion Simulation on the GPU for 3D terrains": https://www.diva-portal.org/smash/get/diva2:1646074/FULLTEXT01.pdf
• "Large Scale Terrain Generation from Tectonic Uplift and Fluvial Erosion": https://inria.hal.science/hal-01262376/document
• "Procedural Modeling of the Great Barrier Reef": https://easychair.org/publications/preprint_download/qLmc
• "Desertscape Simulation": https://www.researchgate.net/profile/Axel-Paris/publication/335488341_Desertscape_Simulation/links/5db7f667a6fdcc2128e8d1d9/Desertscape-Simulation.pdf Adds a sand layer and a vegetation layer. Reproduces realistic dune formations.
• "Procedural Generation of Large-Scale Forests Using a Graph-Based Neutral Landscape Model": https://media.proquest.com/media/hms/PFT/1/Ijdn4?_s=ANyDa8athS4X9z8tNeVbdzsdiSQ%3D First skim: Segments a grip-based landscape into patches of wood species growth.
• "AutoBiomes: procedural generation of multi-biome landscapes": https://d-nb.info/121797170X/34 Uses a climate simulation, then derives biomes from the results.
• "Efficient Animation of Water Flow on Irregular Terrains": http://www-cg.cis.iwate-u.ac.jp/lab/graphite06.pdf
• "Forming Terrains by Glacial Erosion": https://inria.hal.science/hal-04090644/file/Sigg23_Glacial_Erosion__author.pdf
• "Interactive Generation of Time-evolving, Snow-Covered Landscapes with Avalanches": https://inria.hal.science/hal-01736971/file/interactive-generation-time.pdf
• "Waterfall Simulation with Spray Cloud in different Environments": https://www.jstage.jst.go.jp/article/artsci/15/3/15_111/_pdf
• "Interactive Procedural Modelling of Coherent Waterfall Scenes": https://inria.hal.science/hal-01095858/file/waterfall.pdf
• "Slab and Powder-Snow avalanche animation on the GPU": https://bth.diva-portal.org/smash/get/diva2:1606730/FULLTEXT02.pdf