SmartGripperHackathon

Project Structure Overview

• /src Contains the core implementation code that is deployed to and executed on the Gripper.

• /gui Provides an interactive graphical interface. An additional README.md file within this folder explains how to use the GUI.

• /docs Includes all reference materials used during development. Full books are provided to ensure consistency and completeness in documenting the model built in MATLAB.

• /object_classification Contains code for object classification, which can be integrated with the GUI to enhance the overall user experience.

Details regarding the simulation of the model are provided in the following section.

---

• Simulations and Modeling

  In this folder we :

  • Simulate the desired force we need to apply in a object through a fuzzy logic decision

    This script builds, visualises, and tests a Fuzzy-Inference System (FIS) called GripperFIS. The controller estimates how much gripping force a robotic gripper should apply, given two object properties:

    Variable Role Range Linguistic terms (bell-shaped MFs)

    - Weight	Input	0 – 500 g	VeryLight · Light · Medium · Heavy
    
    - Hardness	Input	0 – 1 (soft → hard)	VerySoft · Soft · Hard · VeryHard
    
    - Force	Output	0 – 10 (arbitrary units)	VeryLow · Low · Medium · High
    

    1 . Define inputs and outputs addInput and addOutput create the three FIS variables. Each variable is assigned 4 generalized bell membership functions (gbellmf) that overlap smoothly and look like Gaussian “bells”.

    fis = addInput(fis,[0 500],'Name','Weight');
    fis = addMF(fis,'Weight','gbellmf',[40 2 50],'Name','VeryLight');
    ...
    fis = addInput(fis,[0 1],'Name','Hardness');
    ...
    fis = addOutput(fis,[0 10],'Name','Force');
    Parameter vector [a b c] means:
    a = width, b = slope, c = centre of the bell.
    
    2 . Create the rule base (4 × 4 = 16 rules)
      ruleList maps every combination of the two inputs to a force level.
    
      Example rule:
    
          If Weight is Heavy and Hardness is Soft then Force is High
      Rules are imported with addRule(fis,ruleList);.
    
    3 . Visualise the system
      plotfis shows a block diagram of the FIS.
    
      plotmf plots the membership functions for each variable.
    
      The surface plot (surf) illustrates how Force varies over the full Weight–Hardness grid.
    
    4 . Quick test case
      evalfis(fis,[300 0.4]) evaluates the controller for a 300 g object with medium hardness (0.4) and prints the recommended force.
    

  • model the BLDC motor system we got from the documentation and we apply an adaptive pid controller for the control

    ⚙️ Adaptive PID Controller with BLDC Motor Simulation and Wiener Filtering This MATLAB script simulates an adaptive PID control loop for a Brushless DC (BLDC) motor system. It includes:

    A 2st-order electromechanical model of a BLDC motor
    
    A PID controller that adapts its gains using gradient descent
    
    A custom calculation of settling time to evaluate controller performance
    
    A Wiener filter for post-processing noisy output data
    
    📌 Features  
    🔧 PID Control with Online Tuning
    The PID controller updates its parameters (Kp, Ki, Kd) at each time step based on the current error and the error in settling time, aiming to:
    
    Track a given target speed
    
    Achieve a desired settling time
    
    🔄 BLDC Motor Model
    The model consists of coupled electrical and mechanical dynamics:
    
    
    📉 Wiener Filter for Signal Denoising
    After simulation, the noisy output (motor speed) is passed through a Wiener filter for noise suppression.
    
    📈 Output
    Real-time PID gain updates printed in the console
    
    Plots:
    
      System response with adaptive PID tuning
    
      Denoised output using Wiener filtering
    
    Example:
    
      t=3.50, omega=5.42, settling=1.02, Kp=98.75, Ki=48.75, Kd=0.08
    
    📦 Main Components
    
    Section	Description
    
    for loop	Main simulation loop
    
    calculateSettlingTime()	Computes time until the output stabilizes within ±5% of target
    gradient descent	Updates PID gains based on combined error
    wiener2()	Applies Wiener filter to noisy signal
    plot()	Shows system response and filtering results
    
    🧠 Learning Logic
    
    The learning step adjusts all PID gains equally using:
    
    grad = 0.1 * (tracking_error + settling_time_error);
    Kp = Kp - lr * grad;
    
    ⚠️ Note: As we will see we achieved an optional pid for the dynamic system through control and learning for the parameters
    

🛠️ System Parameters

Symbol Description Value R Motor resistance 50 Ω L Motor inductance 0.1 H Ke Back EMF constant 1 V·s/rad Kt Torque constant 1 Nm/A J Rotor inertia 0.01 kg·m² B Viscous friction 0.001 N·m·s

📊 Sample Visualizations

✅ System response vs. target

🔵 Filtered signal using Wiener filter