fuzzy-yaml

Fuzzy logic control C++ library loaded from a YAML config file

This is a refactored version of the source code originally used in the following paper

B. P. Putra, G. S. Mahardika, M. Faris, and A. I. Cahyadi, "Humanoid Robot Pitch Axis Stabilization using Linear Quadratic Regulator with Fuzzy Logic and Capture Point", arXiv [cs.RO]. 2021.

Instead of using arrays for storing the linguistic sets, key-value maps are utilized for a more natural way of accessing them by calling their respective string key.

Note: As of May 2024, this library is only applicable for the Mamdani type-1 fuzzy inference system with:

• trapezoid and triangle membership functions, and
• 2-dimensional rule table with AND operation

However, contributions for additional features are welcome.

Dependencies

1. CMake

2. Boost

3. yaml-cpp

Build & Run

1. Make sure you have installed the dependencies.

2. Open a new command line terminal. Clone the repository

   git clone https://github.com/BagaskaraPutra/fuzzy-yaml.git

3. Build the project using CMake.

   cd fuzzy-yaml
   mkdir build
   cd build
   cmake ..
   make

4. The example executable files will be generated in the build/examples/ directory.

5. Run the executable, for example:

   ./examples/multiAnklePitch

6. If you wish to run other YAML config files, add the file path after the executable. For example:

   ./examples/anklePitch ../config/allTrapezoidAnklePitch.yaml

Tutorial

A typical YAML config file for the fuzzy-yaml library consists of three main nodes:

inputVariables:
  ...
outputVariables:
  ...
ruleBase:
  ...

The inputVariables node is structured as:

inputVariables:
  Crisp input variable 1:
    memberships:
      Linguistic fuzzy input variable 1.1:
        type: ...
        ...
      Linguistic fuzzy input variable 1.2:
        type: ...
        ...
  Crisp input variable 2:
    memberships:
      Linguistic fuzzy input variable 2.1:
        type: ...
        ...
      Linguistic fuzzy input variable 2.2:
        type: ...
        ...
    ...

Similarly for the outputVariables node:

outputVariables:
  Crisp output variable 1:
    memberships:
      Linguistic fuzzy output variable 1.1:
        type: ...
        ...
    defuzzifier: ...
  Crisp output variable 2:
    memberships:
      Linguistic fuzzy output variable 2.1:
        type: ...
        ...
    defuzzifier: ...
  ...

For the outputVariables node, we have an additional defuzzifier method parameter. As of now, only the centroid defuzzifier method is available.

On the other hand, the ruleBase node consists of the following structure:

ruleBase:
  Crisp output variable 1:
    # rules
    ...
  Crisp output variable 2:
    # rules
    ...

For now, the ruleBase only accepts the 2-dimensional table format. For example in the examples/anklePitch.yaml file, the rule table

ruleBase:
  Kp:
    row_variable: Angle
    row_values: [NegativeHigh, Negative, Positive, PositiveHigh]
    column_variable: AngularVelocity
    column_values: [Negative, Average, Positive]
    operator: AND
    table: 
      - [High, High, High]
      - [Medium, Medium, Medium]
      - [Zero, Zero, Zero]
      - [High, High, High]

is equivalent to having the following 12 fuzzy rules:

• IF Angle is NegativeHigh AND AngularVelocity is Negative THEN Kp is High
• IF Angle is Negative AND AngularVelocity is Negative THEN Kp is Medium
• IF Angle is Positive AND AngularVelocity is Negative THEN Kp is Zero
• ...
• IF Angle is PositiveHigh AND AngularVelocity is Positive THEN Kp is High

Citation

If you intend to use this code in your work, you may cite it as:

@misc{putra2021humanoid,
      title={Humanoid Robot Pitch Axis Stabilization using Linear Quadratic Regulator with Fuzzy Logic and Capture Point}, 
      author={Bagaskara Primastya Putra and Gabrielle Satya Mahardika and Muhammad Faris and Adha Imam Cahyadi},
      year={2021},
      eprint={2012.10867},
      archivePrefix={arXiv},
      primaryClass={cs.RO}
}