a-b-street/abstreet
1
1
Fork 0
mirror of https://github.com/a-b-street/abstreet.git synced 2024-12-29 17:34:58 +03:00
Code Issues Projects Releases Wiki Activity
71be19163d
abstreet/docs/TODO_quality.md
Dustin Carlino 6e2689f68c a round of adding/removing things from the TODOs
2019-04-30 15:59:34 -07:00

6.2 KiB
Raw Blame History

TODO - Map quality

Boundary clipping

  • lakes missing from small_seattle
    • just connecting the ends of ways doesnt always work well
    • maybe increase the Bounds for areas, and let clipping clean up later?
  • some border intersections have weird OOBish geometry, or the arrows look weird
  • simplify border node detection, only do it in convert_osm?

Geometry

  • bad polyline shifting remains

    • from the remaining cases, looks like we need to totally remove some tight points and retry
    • make polygons use the corrections too?
    • bad polyline shifting causes jagged lane endings in generalized_trim_back

  • handle small roads again somehow?

    • reduce degenerate min trim. the disabled fix doesn't look great.
    • I40, I25, I0 cut corners when merged. disabled fix works, but breaks other things.
    • try it bigger
    • deal with loop roads?
    • model U-turns

  • degenerate-2's should only have one crosswalk

    • then make them thinner

  • car turns often clip sidewalk corners now

  • draw SharedSidewalkCorners just around the ped path, not arbitrarily thick

    • dont forget to draw the notches

  • figure out what to do about yellow center lines

    • intersections on one-ways look weird
    • yellow and white lines intersect cars and turn icons and such
    • who should own drawing them?

More data

Low-priority geometry issues

More problems to fix

  • Disconnected map
    • now that LCing model is simple...
  • Impossible turns (from a far bus lane to a crazy left)
  • Buildings intersecting roads, probably because bad lane inference
    • when this happens, get rid of parking lanes first (one or both sides?)
    • iterative process... have to redo affected roads and intersections
    • we havent filtered buildings by proximity to sidewalk yet
      • if we dont filter at all, we pick up some houseboats! :) should draw water...
  • undo disabled traffic signal assertion

Map edits

  • lane type can affect border intersections
  • lane type can affect turn idx
    • assert turns are the same
  • crash when loading some edits (priori->bikelanez)
    • edit an intersection first, then change lanes to destroy turns. need to recalculate the policy, preserving edits!
      • just revert intersection and warn
      • or store overrides more granularly and warn or do something reasonable

Release

  • publish the map data

Sim bugs/tests needed

  • do bikes use bike lanes?
  • test that peds will use buses organically
    • make sure that we can jump to a ped on a bus and see the bus
  • park/unpark needs to jump two lanes in the case of crossing a bike lane or something
    • should only be able to park from the closest lane, though!

Discrete-event sim model

  • gridlock

    • get_car_positions will recurse?
    • laggy head stuff will be inefficient without prevention?
    • preventish... dont start turn unless target WILL have capacity
      • stronger is room_at_end now, but thats too strict
      • best-case static capacity wont help (we might have longer buses and over-commit)
      • worst-case static capacity will stop shorter cars from going when they legit could
      • look at current occupants, find where the back WILL eventually wind up (just add up lengths + following dist, actually), then reserve after that.

  • cleanup after the cutover

    • explicit tests making cars park at 0 and max_dist, peds walk to 0 and max_dist
    • proper intersection policies, by seeing full view
    • time travel mode can be very smart
    • dupe code for initially spawning vs spawning when a trip leg starts.
      • validation (just do it in spawn and return an error?)
      • getting the path
      • converting goal into router and such
    • is overtime necessary?
    • all the scoring/query stuff is so terrible

  • perf

    • speed up Scheduler

      • BinaryHeap
        • binheapplus not serde
        • std::cmp::Reverse not serde
        • so have to manually impl ord
        • BinaryHeap isnt PartialEq, which sim det stuff needs!
        • update() is very hard
      • priority_queue crate... internally uses hash, so serialization and determinism probably borked

    • dig into individual events, still too many?

      • for laggy heads, often round down and try slightly too early

    • change sim.step API to something like step_to_time

      • handle savestating better... use scheduler, but ensure at the end of that time
      • make the desired speed in editor work better