A sandbox practice for the particle swarm optimization.
Use PSO to find the best the RBFN parameters, which input is the distances detected from car radar and output is the angle of wheel.
Minimize the err:
err = (1 / N) * sum(abs(d.expected_output - RBFN(d.input)) for d in training_dataset)N is the total number of training data.
On the other word, we want to maximize the fitness value (1 / err ).
The position and velocity for each individual in population of PSO is updated by following formula.
- Velocity
self.velocity = inertia_weight * self.velocity + \
cognitive_const * (self.best_position - self.position) + \
social_const * (global_best_position - self.position)
# limit the velocity
np.clip(self.velocity, -self.v_max, self.v_max, out=self.velocity)
self.position += self.velocity
# limit the position to fit the range of RBFN parameters
rbfn_params_limiter(self.position)- Download this project
git clone https://gitlab.com/seanwu1105/pso-car.git- Change directory to the root of the project
cd pso-car/- Run with Python interpreter
python3 main.py| Input (Distances) | Output (Wheel Angle) |
|---|---|
22.0000000 8.4852814 8.4852814 |
-16.0709664 |
# Front_Distance Right_Distance Left_Distance Wheel_Angle
22.0000000 8.4852814 8.4852814 -16.0709664
21.1292288 9.3920089 7.7989045 -14.7971418
20.3973643 24.4555821 7.2000902 16.2304876
19.1995799 25.0357595 7.5129743 16.0825385
18.1744869 42.5622911 8.0705896 15.5075777The data location is /data. The application will load every files with *.txt extension automatically after the execution.
0,0,90 // the starting position and angle of car (x, y, degree)
18,40 // the top-left coordinate of the ending area
30,37 // the bottom-right coordinate of the ending area
-6,-3 // the first point for the wall in map
-6,22
18,22
18,50
30,50
30,10
6,10
6,-6
-6,-3 // the last point for the wall in mapEvery coordinates between the fourth and last line are the corner point of the walls in map.
pip3 install numpypip3 install matplotlibpip3 install pyqt5pip3 install pyside2