FastSLAM

Python simulation of FastSLAM

Intall Dependencies

Using a new virtual env to install the packages: pip install -r requirements.txt

Run Simulation

1. Run FastSLAM 1.0 python fast_slam.py

2. Run FastSLAM 2.0

Control

Using arrow keys to control the robot, you can set number of steps in fast_slam.py.

Sensor

Currently, there are 4 landmarks in the world. You can add more landmarks in the world.py by modifying setup_world method. The coordinates are using the bottom-left corner point as the origin.

In the sense method in the particle.py, the robot randomly observe 2 landmarks and measure the distance and the direction to the landmarks. Then it adds the Gaussian noise to the measurements. The noise level is set up in the set_noise method. Only robot has the bearing_noise:measurement errors for the angles, and distance_noise: measurement errors for the distance. You can also set the motion noise for the robot and particles.

The obs_noise is the additive part of the prediction step of the EKF. First term specifies the error for distance and the second term specify the error for angles. Larger the value, more relax the model will be when considering the data association. obs_noise should be at the same magnitude as distance_noise and bearing_noise.

The control_noise attribute model the motion noise. First two terms specify the error for the x, y coordinates and the third term for the orientation.

Souce

fastSLAM paper