In this project, along with the class of fuzzy controllers used in the simulator, the classes related to the membership in each of the fuzzy sets of proximity to the left, right, center(accross), wheel rotation, and gas pedal are also defined.
membership , fuzzify , and defuzzify all have common implementations in their parent class. In the Right, Left and Rotate classes, in the __init__ method, only the set parameters including the set labels and membership functions
of each of them are stored in a dictionary in the parameters variable and the names of the sets (labels) are stored in the fuzzy_labels variable of the class.
Finally, the fuzzify and defuzzify methods of the controller use the methods of the belonging classes and the main controller is not directly involved in their calculations.
The implementation of the inference method is general purpose and completely placed in the Utils class. The decide method is also used as the interface used in the simulator.py.
In addition to inference, the Utils class contains functions that read the rules and parse them for use. The structure of the speed fuzzy controller is similar to the steering wheel rotation controller.
In the decide method, which is actually the interface between the controller and the simulator, first the absolute distances from the left and right are given as input to the method, and finally an absolute output is given to the simulator to rotate the steering wheel.
Bsides, in the decide method, the first task is to fuzzify the input values, which is done by the fuzzify method of the controller, and the values of the distance from the left and the right are grouped in three labels/groups, close, balanced, and far, and returned to decide.
Then the inference method, according to the rules that were loaded from the file during the construction of the controller with the help of the parse_rule and read_rules method,
- based on the membership status of the antecedent combination,
- based on the AND and OR operators,
- and the percentage of consequent activity (if several different rules lead to activation, they share consequences, the level of activation of the consequence of the rule with the highest percentage of activation) is return by the method.
Finally, in the defuzzify method, by approximately calculating the center of mass of the output parameter, i.e. the rotation of the command and gas pedal, it calculates the center of mass of the active area and returns the absolute value through its defuzzify call.
One of the important parameters of this step is the integration delta, which is effective in the accuracy of center of mass approximation and finally decision making.
run python simulator.py in the terminal