Skip to content

Latest commit

 

History

History
73 lines (51 loc) · 2.44 KB

TODO.md

File metadata and controls

73 lines (51 loc) · 2.44 KB
Raw

Future work

Simulation

Implementation improvements

  • Think of better name for project!
  • Make simulation frame independent (fix your time step article is great)
  • Use sentinel values in spatial grid to allow exactly the number of requested boids.
  • Speed up computation with a parallel collection library like Rayon.
  • Dynamically select correct shell gap starting size.
  • Sort the neighbourhood lookup arrays into memory access pattern order.
  • Really dig down into runtime perf - use testing tools to find hotspots, bad caching
  • Dont recalculate the forces of each boid every frame, just enough of them. (expose the amount/strategy for doing this to the user?)
  • Partition the boids and spread the update calculation over several frames.

Ideas

Up next

  • Have several presets for different kinds of flock
  • Option to support cursor interaction only when pressing down the mouse button.

Maybe one day

  • Support automatically reloading config file when it changes.
  • Dynamically calculate pleasing default parameters based on window size and resolution.
  • Further explore the feel of the simulation.
    • Different sized neighbourhood lookup table patterns.
    • Can we detect how busy the neighbourhood is and use it to scale repulsion, based only on some immediate/sampled neighbours positions?
      • Could such a "panic factor" overcome MAX_FORCE? Have a dynamic max force?
      • You could infer than an area is busy from extreme closeness
      • Maybe we can take a cue from reynolds subsumption architecture? Disable one behaviour in favour of another?
    • Allow user to tradeoff between perf and accuracy.
    • Throw in randomness or bias to partial sorting algorithm
    • When things are busy/crowded/"angry":
      • Use a dynamic neighbourhood range (don't need big range for "calm" flock)
      • Sample neighbourhood
      • Add a random "panic" force

Renderer

Implementation improvements

  • Handle resizing of screen.

Ideas

Up Next

  • Offer more than one renderer/shader.

Maybe one day

  • Support switching between different renderers/shaders at run time.
  • Allow running at different resolutions.

Fps Counter

  • Rethink how caching works, maybe this doesn't live in fps module.
  • Consider building an ring buffer of Instant instead of Duration

Main / Application

Implementation improvements

  • Dont rely on vsync to keep 60fps
  • Update glutin
  • Slim down size of main loop