abstreet/docs/TODO_phase3.md

58 lines
2.3 KiB
Markdown
Raw Normal View History

2018-07-07 00:53:22 +03:00
# TODO for Phase 3 (Simulation)
2018-07-09 00:04:19 +03:00
## cars
2018-07-07 00:53:22 +03:00
2018-09-17 02:04:33 +03:00
- features
- different lookahead reaction times
- figure out lanechanging (and turns that dont make sense)
- code cleanup
- figure out responsibility btwn agents and managers, then fix up visibility
- things like ParkingSimState have so many methods -- some are only
meant for spawner, or driving/walking to query. separate out some
traits.
- on a lane vs turn permeates so many places
2018-07-09 00:04:19 +03:00
2018-07-07 00:53:22 +03:00
- better visualization
- draw moving / blocked colors (gradually more red as they wait longer)
2018-08-09 23:01:47 +03:00
- make lookahead buffer follow the shape of the road and extend into other lanes and stuff
2018-09-17 02:04:33 +03:00
- render cars with textures?
2018-07-07 00:53:22 +03:00
- reversible sim
2018-09-05 22:37:57 +03:00
- be careful
- 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.
2018-09-06 00:36:10 +03:00
- no way for an agent to request a turn and ASAP have it granted. are there cases where they might slow down unnecessarily?
2018-09-05 22:37:57 +03:00
2018-07-09 00:04:19 +03:00
## bikes
2018-07-07 00:53:22 +03:00
2018-07-23 20:50:01 +03:00
- model bikes as slow cars
2018-07-07 00:53:22 +03:00
2018-07-09 00:04:19 +03:00
## pedestrians
2018-08-01 00:13:15 +03:00
- render overlapping peds reasonably
## General
- savestating a sim has nondet output due to hashes; switching to btree is kind of weird
- consider overriding encoding for TurnID and such, instead of remembering to stick maps everywhere
- diffing two sim states is tedious no matter what; is there a nice macro-driven deep equals we could do instead?
- will need programmatic diffs later for pointing out changes to players in A/B tests
- consider refactoring car/ped sim
- basic structure with actions, react, stepping is same. SimQueue, lookahead, can goto? differs.
2018-09-17 02:04:33 +03:00
- handle map borders
- manually mark polygon, not a bbox, as the border
2018-09-17 02:04:33 +03:00
- be able to start/end trips there
- start to model demand data with manually selecting a source/dest area
- figure out how to capture stacktraces kinda optionally
- manual call at a fxn to dump its stacktrace somewhere (to a file? ideally shared global state to dedupe stuff)
- macro to insert a call at the beginning of a fxn
- macro to apply a macro to all fxns in an impl
- then i can manually edit a few places when I want to gather data
- https://en.wikipedia.org/wiki/File:A_Call_Graph_generated_by_pycallgraph.png