Config Registry

The config module stores references to classes and config dataclasses. It’s used internally to know how to create many types of objects in the simulator:

• swarmsim.agents

  • swarmsim.agent.controllers

• swarmsim.worlds

  • swarmsim.world.objects

  • swarmsim.world.spawners

• swarmsim.sensors

• swarmsim.metrics

This allows you to register your own class definitions which will work in .yaml files.

See also

See the swarmsim.config module for the API.

See the Our YAML module for information on RobotSwarmSimulator’s custom .yaml parser and tags

agent and world Configs

RobotSwarmSimulator uses dataclasses to define the configurations for agents and worlds. These dataclasses are used to create the corresponding objects in the simulator.

For example, let’s create a new agent class called MyAgent that has a MyAgentConfig dataclass.

MyAgent.py

from dataclasses import dataclass
from swarmsim.config import associated_type, filter_unexpected_fields
from swarmsim.agent.BaseAgent import Agent, BaseAgentConfig

@associated_type("MyAgent")
@filter_unexpected_fields
@dataclass
class MyAgentConfig(BaseAgentConfig):
    my_custom_field: int = 999

class MyAgent(Agent):
    pass

The associated_type() decorator associates the MyAgentConfig dataclass with the MyAgent class by adding a config.type = 'MyAgent' field to the dataclass.

test_custom_agent.py

from swarmsim.config import register_agent_type
from MyAgent import MyAgent, MyAgentConfig

register_agent_type('MyAgent', MyAgent, MyAgentConfig)

Once your agent class is registered with the config system, you can load a .yaml file with a type: MyAgent field, RectangularWorld will know how to create a MyAgentConfig from your .yaml and subsequently create an instance of MyAgent.

world.yaml

type: "RectangularWorld"
agents:
  - type: MyAgent  # this becomes MyAgentConfig
    my_custom_field: 100

A similar system is used for World types and their associated config classes, but there’s currently only one world type: RectangularWorld

Note that objects are a special type of agent , so they also use this system.

Everything Else (dict-like config objects)

For everything that isn’t an agent or world , the config system doesn’t use dataclasses. Instead, it uses a dictionary-like object that has a type field. This includes everything from spawners to metrics and control .

For example, the StaticController has a type field that is used to determine how to create the controller.

SpinningController.py

from swarmsim.agent.control.Controller import AbstractController

class SpinningController(AbstractController):
    def __init__(self, parent,
       angular_velocity: float,
    ):
        super().__init__(parent)
        self.angular_velocity = angular_velocity

    def get_actions(self, agent):
        return 0, self.angular_velocity

Then, register the controller with the config system:

test_custom_controller.py

from swarmsim.config import register_dictlike_type
from SpinningController import SpinningController

register_dictlike_type('controller', 'SpinningController', SpinningController)

And then you can use it in a .yaml file:

world.yaml

type: "RectangularWorld"
agents:
  - type: MyAgent
    controller:
      type: SpinningController
      angular_velocity: 0.1  # rad/s