Backend Client Architecture

To maintain consistency from one backend client to another and to promote modularity, we make use of several interfaces. Any new backend client should inherit from ClientInterface and make use of the PlannerInterface. Methods for connecting, disconnecting, and generally managing the client state are a part of the client, while any methods for planning, scene management or kinematics are attached to the planner. Eventually, methods for execution and control will be included in the ControlInterface, but for now, such methods and attributes will be left with the client.

The PlannerInterface serves as a template for any client-specific planner, providing default behavior for each of the methods listed within. When a developer wishes to override any of these defaults, they should make use of the appropriate backend feature interface from backends/interfaces.py. The file interfaces.py consists of a collection of classes, any implementation of which is callable through its __call__ magic method. For example:

from compas.geometry import Frame
from compas_fab.backends.interfaces import InverseKinematics

class ExampleInverseKinematics(InverseKinematics):
    def inverse_kinematics(self, robot,
                           frame_WCF,
                           start_configuration=None,
                           group=None,
                           options=None):
        # insert fancy code here
        pass

can be instantiated and called in the following manner:

calculate_example_ik = ExampleInverseKinematics()
frame = Frame([0, 0, 0], [1, 0, 0], [0, 1, 0])
ik_result = calculate_example_ik(robot, frame)
# or equivalently:
ik_result = calculate_example_ik.inverse_kinematics(robot, frame)

These backend feature interfaces exist in part to enforce a common signature across all implementations of, say, inverse_kinematics for greater end-user ease. Please adhere to the types listed for the arguments and return values listed in the documentation for the backend features as much as possible.

These interfaces as exist to allow mixing and matching of the backend features of various clients to suit the performance and overhead requirements of the end-user. To illustrate this last point, consider the following example, where the backend of ClientA is very efficient at computing inverse kinematics and has no feature to plan motion, while the backend of ClientB is slow to compute inverse kinematics but can plan motion:

with ClientA() as client_a, ClientB() as client_b:
    inverse_kinematics = ClientAInverseKinematics(client_a)
    plan_motion = ClientBPlanMotion(client_b)

Here we can assign the inverse kinematics to be calculated by the backend of ClientA, while the motion planning is calculated by the backend of ClientB. (We assume ClientA and ClientB inherit from ClientInterface and that ClientAInverseKinematics and ClientBPlanMotion inherit from InverseKinematics and PlanMotion, resp.)