Dl For Navigation

Implementation of Regression Models on Navigation with IMUs.
Alternatives To Dl For Navigation
Project NameStarsDownloadsRepos Using ThisPackages Using ThisMost Recent CommitTotal ReleasesLatest ReleaseOpen IssuesLicenseLanguage
24 days ago2gpl-3.0C++
Multi-Sensor Fusion (GNSS, IMU, Camera) 多源多传感器融合定位 GPS/INS组合导航 PPP/INS紧组合
5 months ago9otherMATLAB
NaveGo: an open-source MATLAB/GNU Octave toolbox for processing integrated navigation systems and performing inertial sensors analysis.
4 years ago3mitC++
10 days ago42mitJulia
Robot Motion Estimate: Tools, Variables, and Factors for SLAM in robotics; also see Caesar.jl.
3 months agobsd-3-clauseJupyter Notebook
Neural-Kalman GNSS/INS Navigation for Precision Agriculture
Platform Aceinna_imu31
10 months ago12Python
Aceinna Navigation Studio: open-source, embedded development platform for Aceinna IMU hardware. Run custom algorithms and navigation code on Aceinna IMU/INS hardware
3 months ago15Jupyter Notebook
A lidar-based Teach-and-Repeat framework designed to enable outdoor autonomous navigation in harsh weather.
3 years ago
|| A ROS package of the WaveShare Jetbot AI Kit. A Differential Drive Robot ||
a day agobsd-3-clausePython
A differential drive robot is controlled using ROS2 Humble running on a Raspberry Pi 4 (running Ubuntu server 22.04). The vehicle is equipped with a raspberry pi camera for visual feedback and an RPLIDAR A1 sensor used for Simultaneous Localization and Mapping (SLAM), autonomous navigation and obstacle avoidance.
Rins W9
3 years agomitPython
RINS-W: Robust Inertial Navigation System on Wheels
Alternatives To Dl For Navigation
Select To Compare

Alternative Project Comparisons

Implementation of Regression Models(DRNN + SVR + KNR) Against Kalman Filters for the Calculation of Quaternions and Euler Angles

This Repo Contains 2 Main Parts:

DataSet Collection

Data Contains

1. Gyroscope Data 3DoF
2. Accelerometer Data 3DoF
3. Magnetometer Data 3DoF
4. Euler Angles Data 3DoF
5. Quaternions Data 4DoF

To collect the data, I made a small setup using a Raspberry pi and BNO055 IMU. So simple but it works wonderfully.

1. Hardware Setup

1. Raspberry pi 4
2. BNO055 9DoF IMU
3. Lithium Polymer Battery
4. Buck Converter
5. USB-C to USB-A Adapter
6. Breadboard
7. Bunch of Jumper Wires
**Note:** It should be portable, So the data collection is not constrained to a specific location and angle.

Best Data Logger in the Whole World!

2. Software Setup

Here is the code for the data collection. First, the code is written in C++. The code contains a loop with a constant frequency of 100Hz. The loop reads the IMU data for 5 minutes and writes it to a ".csv" file at each iteration.

There is a Executable file for the Code in the same folder. But in case you want to run the code, you need to install the WiringPi library from here.

There is another prerequisite for the code to run. High speed I2C (400kHz) in the used Raspberry pi 4 have to be anabled. For the purpose use the procedure described in the here.

After handling all, the code is ready to be compiled.

So, Navigate to the folder where the code is and run the following bash command:

$ g++ -o CollectorApp *.cpp -lwiringPi -lm -std=c++11

Now the executable file is ready to be run like this:

$ ./CollectorApp

There are 8 Available Files Containing Data in the "DataSet" Folder.

Note: The files having "(No Mag)" in there names, don't contain magnetometer data.


For the purpose of this project, I used the following models:

  • Deep-RNN Model
  • Support-Vector-Regression Model
  • K-Neighbours-Regression Model

Seems Like the Complexity of the data is so much that the RNN Model is not able to handle it.

And it had a huge RMS of 0.5 in it's predictions and also the model stopped learning after a few epochs. Although the learning rate was adaptively adjusted.

So, I decided to use the Multi Output Support Vector Regression model. Which seemed to handle the data better. But, also the Model didn't perform well enough!

So, as a last resort, I used the K-Neighbours Regression model. Which turned out to be the finest model for this data...

And with a bit of tuning, it was able to perform the best on the data. with a mean absolute error of 0.0043 for Quaternions and 3.3 for Euler Angles.

(Note: The mean absolute error is the average of the absolute errors of the predictions)

Logs for all the models are saved in a file named "Logs.txt" in the specific folder of each model.

Popular Imu Projects
Popular Navigation Projects
Popular Hardware Categories
Related Searches

Get A Weekly Email With Trending Projects For These Categories
No Spam. Unsubscribe easily at any time.
C Plus Plus
Deep Learning
Raspberry Pi
Neural Network
Recurrent Neural Networks