API
Suppose you're tired of manually fiddling with traffic signals, and you want to use machine learning to do it. You can run A/B Street without graphics and automatically control it through an API.
Examples
This Python example has everything you need to get started.
Also check out the Go example, which demonstrates just a few of the API calls.
API details
Under construction: The API will keep changing. There are no backwards compatibility guarantees yet. Please make sure I know about your project, so I don't break your client code.
For now, the API is JSON over HTTP. The exact format is unspecified, error codes are missing, etc. A summary of the commands available so far:
- /sim
- GET /sim/reset: Reset all map edits and the simulation state. The trips
that will run don't change; they're determined by the scenario file you
initially pass to
headless
. - POST /sim/load: Switch the scenario being simulated. Takes a SimFlags as a JSON POST body.
- GET /sim/get-time: Returns the current simulation time.
- GET /sim/goto-time?t=06:30:00: Simulate until 6:30 AM. If the time you specify is before the current time, you have to call /sim/reset first.
- POST /sim/new-person: The POST body must be an ExternalPerson in JSON format.
- GET /sim/reset: Reset all map edits and the simulation state. The trips
that will run don't change; they're determined by the scenario file you
initially pass to
- /traffic-signals
- GET /traffic-signals/get?id=42: Returns the traffic signal of intersection #42 in JSON.
- POST /traffic-signals/set: The POST body must be a ControlTrafficSignal in JSON format.
- GET /traffic-signals/get-delays?id=42&t1=03:00:00&t2=03:30:00: Returns the delay experienced by every agent passing through intersection #42 from 3am to 3:30, grouped by direction of travel.
- GET /traffic-signals/get-cumulative-thruput?id=42: Returns the number of agents passing through intersection #42 since midnight, grouped by direction of travel.
- /data
- GET /data/get-finished-trips: Returns a JSON list of all finished trips. Each tuple is (time the trip finished in seconds after midnight, trip ID, mode, duration of trip in seconds). The mode is either a string like "Walk" or "Drive", or null if the trip was aborted (due to a simulation bug or disconnected map).
- GET /data/get-agent-positions: Returns a JSON list of all active agents. Vehicle type (or pedestrian), person ID, and position is included.
- /map
- GET /map/get-edits: Returns the current map edits in JSON. You can save
this to a file in
data/player/edits/map_name/
and later use it in-game normally. You can also later run theheadless
server with--edits=name_of_edits
.
- GET /map/get-edits: Returns the current map edits in JSON. You can save
this to a file in
Working with the map model
If you need to deeply inspect the map, you can dump it to JSON:
cargo run --bin dump_map data/system/maps/montlake.bin > montlake.json
The format of the map isn't well-documented yet. See the generated API docs and the map model docs in the meantime.
Working with individual trips
You can use the /sim/new-person API in the middle of a simulation, if needed. If possible, it's simpler to create a Scenario as input.
Working with Scenarios
You can import trips from your own data.
You can also generate different variations of one of the demand models by specifying an RNG seed:
cargo run --bin random_scenario -- --rng=123 --map=data/system/maps/montlake.bin > data/system/scenarios/montlake/home_to_work.json
You can also dump Scenarios (the file that defines all of the people and trips) to JSON:
cargo run --bin dump_scenario data/system/scenarios/montlake/weekday.bin > montlake_weekday.json
You can modify the JSON, then put the file back in the appropriate directory and use it in-game:
cargo run --bin game data/system/scenarios/montlake/modified_scenario.json
The Scenario format is also undocumented, but see the generated API docs anyway.