Mostly reflective of the content of chapter 3 of Probabilistic Robotics by Sebastian Thrun
Toy kalman filter in 1 dimension
l, r to move the robot (stay within 0,100) m to perform a measurement update
i to re initialize the program q to quit
unfortunately these must be entered in the REPL.
This simplistic robot has two controls - "l" and "r" When either is executed, the robot attempts to move 5 units in the corresponding direction. The actual amount moved is sampled from the ground truth distribution motion covariance Q.
The underlying motion model is sampled as a normal distribution Note that in this case that is correct-- the underlying motion model is normally distributed
It has estimated motion covariance Q.
Robot samples actual distance Robot ground truth moves that distance
Kalman Filter estimate moves that distance, and the covariance is increased accordingly.
Generates the plots from chapter 3.3 on the EKF
Can either use a linear transform, or input your own non-linear function Unfortunately this has to be done by editing the file manually
Based on Chapter 5.3, Velocity Motion Model
Modify the alpha variables to see their effects on the resulting distribution. The first two effect the impact of velocity and angular velocity on the velocity error The second two impact the angular velocity error The final pair impacts the resulting pose gamma
You can manually select a control, by modifying the initial control state. After entering "r" into the repl, the plot will refresh with new random controls chosen from a uniform distribution.
r to reload s -> filename to save image q to quit
| Images | Images |
|---|---|
 |
 |
 |
 |
To be honest this never ended up working so just ignore it.
Create a python3 virtual envrironment
python3 -m venv ./venv
source ./venv/bin/activate
Install Required Packages
pip install notebook=6.4.12
Julia Required Packages: Only thing of note is WebSockets v1.5.9 NOT 1.6.0
02S: https://dataset.amtc.cl/index.php/overview/ 02S_lidar.dat.tar.bz2 02S_radar.dat.tar.bz2 02S_timing.dat.tar.bz2