a-b-street/abstreet
1
1
Fork 0
mirror of https://github.com/a-b-street/abstreet.git synced 2024-11-25 03:41:09 +03:00
Code Issues Projects Releases Wiki Activity
e186b24dac
abstreet/docs/TODO_game.md
Dustin Carlino 9d450363ea be able to launch DES model on any lane
2019-02-09 18:38:44 -08:00

2.2 KiB
Raw Blame History

TODO - Core gameplay

  • parking vs bus tutorial level

  • Example use cases

    • maybe dont need simulation at all to play with these
      • just a smooth editing and diffing UI
    • montlake/520 turn restrictions with pedestrian scramble
    • close interior neighborhoods to most cars (except for src/dst), see how traffic restricted to arterials would work
      • puzzle: with only X miles of retained road, where would you leave them? what roads would go away?
    • create a bike network with minimal hills, dedicated roads, minimal crossings

  • easter eggs

    • name agents, with some good names scattered in (Dustin Carlino, Dustin Bikelino, Dustin Buslino...)

The very detailed driving model

  • different lookahead reaction times

  • could see if we ever have a lookahead constraint to deaccel more than what we're capable of. it might mask problems. but since things like accel_to_stop_in_dist don't have a careful notion of how much time will pass, they recommend big rates sometimes.

  • no way for an agent to request a turn and ASAP have it granted. are there cases where they might slow down unnecessarily?

  • lane changing: seattle polite. planned things work. limit to space needed.

A much simpler model

  • render maps as super simple transit network.

  • event based + FSM + radically simpler model. essence of scarcity -- intersections, parking, lane capacity.

  • simpler kinematics: if we recognize we're stopping for an intersection or following a vehicle closely, just set the final state to be perfectly what we desire, rather than solving for the acceleration to achieve that. intent (and capabilities), not mechanics!

  • check out https://github.com/movsim/traffic-simulation-de

  • collapse smaller roads/neighborhoods and just simulate aggregate stats about them

Discrete-event model

  • Make cars cover an entire lane, even when it's short

  • try real distance over time ;) but does it make intersection calculation extremely hard?

  • avoid impossible accel/deaccel

  • Handle lead car being faster

  • The speed stated in the middle of intervals is clearly wrong for the follower car

  • Prototype alternate driving model

    • branch or forked code?
    • keep timesteps or immediately cut over to events?
    • figure out how lookahead/interruptions and replanning work