PSO Follow Line Deep Learning Simulator

A Processing simulator of a differential line-following robot with traditional control (PID), machine learning (recurrent neural networks), and PSO-based weight optimization (Particle Swarm Optimization). Includes a simplified physical model of the robot, virtual sensors, and an interactive interface for parameter visualization and tuning.

Main Features

• Differential-drive robot with dynamic and kinematic model
• Odometry and virtual front/side sensors for line detection
• PID control and Recurrent Neural Network (RNN) control
• In-simulation PSO optimization of neural network weights
• Menu interface for adjusting evolutionary parameters and viewing information
• Optional logging of simulation data into .txt files

---

Requirements

• Processing 4 (Java Mode)
• Processing Sound library (used for audio feedback)
  • Install at: Sketch > Import Library… > Add Library… > search for “Sound” and install “Sound” (Processing Foundation)

Media files (images) are included in the project’s data/ folder.

How to Run

1. Install Processing 4 and the Sound library (see above).
2. Open the project in Processing: File > Open… and select this repository’s directory (the .pde tabs will load automatically).
3. Make sure the image files exist in data/:
   • data/Iron Cup 2019.png
   • data/Ratão 4.0.png
4. Run the sketch (Play button or Ctrl/Cmd + R). The main file is PSOFollowLineDeepLearning04.pde.

Optional (command line): if you have processing-java installed, you can run:

processing-java --sketch="/path/to/this/project" --run

Controls During Simulation

Keyboard:

• S / N: toggle sound on/off
• A / D: toggle trail drawing
• M / U: show/hide multiple “ghost” robots (visual comparison)
• I / O: open PID interface (informational text)
• C: toggle a guiding circle under the mouse

Mouse:

• Right-click: reposition the robot (center)
• Top bar: click to open the menu; hover tabs to display content; in some tabs you can drag sliders to adjust parameters

Menu and Parameters

The top menu has tabs:

• Genetic Algorithm / PSO: displays and allows tuning of evolutionary parameters
  • Population size, mutation rate, and fitness function weights (stability, speed, distance)
• Physics: displays physical parameters (mass, inertia, gravity)
• PID Control: shows current gains of the outer control loop
• Neural Network: shows network input/hidden/output sizes

Notes:

• PSO optimization runs in a background thread (trainNet) during execution and periodically updates the best weights.
• Metrics such as “generation”, “fitness”, etc. are printed to the Processing console and shown in the menu.

Quick Configuration (in the Code)

Main configurable points are in the setup() of PSOFollowLineDeepLearning04.pde:

• Initial PID gains:

  m[0].setPID(200, 0, 8)
  m[0].malha_ext_esq = new Control(200, 0, 8);
  m[0].malha_ext_dir = new Control(200, 0, 8);

• Neural network (sizes):

  int[] deepLearning = {11, 5, 2}; // input, hidden, output
  int[] temporalLayers = {2, 0, 0}; // temporal layers for RNN

• Dynamics/time:

  m[0].setVelMax(3.45); // (m/s)
  m[0].setVariacaoTempo(1000); // (µs per step)

• Evolution algorithm parameters (class DNA in AI.pde):
  • tamPopulacao (population size), mutacao (rate), fitness weights (pesoVel, pesoDist, pesoEstabilidade), time and distance limits.

Project Structure (Main Files)

• PSOFollowLineDeepLearning04.pde: main sketch; loads track, initializes robot, network, PSO, and simulation loop
• AI.pde (DNA class): population evaluation, fitness calculation, and PSO iteration over network weights
• Robot.pde (Movimento class): robot logic, line reading, PID, neural control, odometry, wheel dynamics
• Dynamic.pde: robot physics (simplified), velocities, integration
• Deep_Learning.pde: implementation of Neural_Network and auxiliary matrix operations
• Algebra.pde: matrix/vector utilities
• Heuristic.pde: PSO implementation (population, individuals, pose updates)
• Interface.pde: top menu, tabs, and sliders
• Controles.pde: Control class (controller parameters/state) and logs
• Botoes.pde: button/clickable area components used by the menu
• data/: static assets (images) used by the sketch
• controle/: optional exported logs (velocity, distance, time) when enabled

Logs and Export

When enabled (see Control.save_data), the simulator saves time-series logs to controle/:

• vel_linear.txt
• vel_estrategia.txt
• dist_percorrida.txt
• tempo.txt

Tips

• If the window does not open or sound fails, check if the “Sound” library is installed in Processing 4.
• To speed up the simulation, adjust velTempo in PSOFollowLineDeepLearning04.pde.
• If the robot “disappears” from the track (leaves the image), reposition it with the right mouse button.

License

Distributed under the MIT license. See the LICENSE file for details.

Credits

Author: Jakson Almeida