A Planner

The planner owns the route table: it finds the junctions, groups their entry lanes, and generates the candidate routes traffic can drive. The traffic manager constructs it once per scenario and queries it for every spawn — on both backends — so replacing it changes what demand is.

Kind planner, constructor (world, config), selected by scenario.traffic.planner.type. (Mechanics — discovery, freeze, override, validate — on Replace a Component.)

Worked example

The built-in marl planner exposes its route queries as methods, so the practical route is subclassing it and changing exactly the behaviour you want to vary:

from opencda_marl.api import register_component
from opencda_marl.builtin.planners.marl_planner import MARLPlanner


@register_component("planner", "keep_right")
class KeepRightPlanner(MARLPlanner):
    """Demand routes prefer a right turn wherever the junction has one."""

    def get_random_route(self, junction_id, visualize=False):
        right = self.get_random_route_by_type(junction_id, "right", visualize)
        return right if right is not None else super().get_random_route(
            junction_id, visualize)

scenario:
  traffic:
    planner:
      type: keep_right

A from-scratch planner implements the same query surface the built-in has (get_junctions, get_routes, get_random_route*); start from opencda_marl/builtin/planners/marl_planner.py as the reference.

Verify

Run the SUMO smoke and watch the Traffic view in the monitor — the routes actually driven are the planner's output:

pixi run python opencda.py -t sumo-marl/intersection/route_rule_path_pcmappo --marl