Aleksandar Haber PhD

Two of most valuable robotics, UAV engineering, and GNC skills are sensor fusion and calibration of Inertial Measurement Units (IMUs). I noticed that a lot of people are struggling with this (not only students, but also engineers). The reason for this is that you have to take at least 5-10 classes on control theory, estimation, discrete-time systems, digital signal processing, and rigid body dynamics to properly understand how IMUs should be calibrated and how sensor fusion should be performed. Basically, you need to do a PhD in control theory :). However, you also have to have hands-on experience with electronics, programming, and measurements, which you can only get if you spend several years in a lab.

Consequently, I decided to create a series of tutorials on how to correctly and in a disciplined manner perform accelerometer, gyroscope, and magnetometer sensor fusion and attitude estimation. This is of paramount importance for robotics and GNC. In the first iteration, I implemented a complementary filter for fusing accelerometer and gyroscope measurements (video shows the results). Accelerometer data is passed through a low pass filter and the gyroscope data is passed through a high pass filter. The filters and data collection are running on Arduino, and data is sent through a serial port to a desktop computer running a Python script that is used to visualize the measurements. I am using Adafruit ICM 20948 IMU (not being paid to promote their product), which costs around $20.

Video below shows the first iteration and results that seem promising: https://www.youtube.com/watch?v=hR9FF...

3 months ago (edited) | [YT] | 8

Aleksandar Haber PhD

News:
- I am writing a book on Robot Operating System 2 (ROS2) Jazzy.
- A significant number of people asked about tutoring and professional engineering training on ROS2, robotics, control, ML, and AI. Currently, I can accept several new students or engineers interested in training and tutoring related to ROS2, robotics, and control. Due to economic reasons and to make sure that I can at least partially compensate for my time, the rates are designed for the people in the US and the US market (in some cases, I can accept students outside of the US). If you are interested, the contact email is: ml.mecheng@gmail.com

Also, in the future, there will be a live course on building a mobile robot from scratch and implementing ROS2 control, navigation and SLAM algorithms. If you are interested, send an email.

6 months ago (edited) | [YT] | 64

Aleksandar Haber PhD

Here is an almost 90 minute long tutorial on how to correctly and in a disciplined manner implement a position controller for a mobile robot in C++ and ROS2 Jazzy Jalisco from scratch:
https://www.youtube.com/watch?v=nWvDr...

7 months ago | [YT] | 13

Aleksandar Haber PhD

Raspberry Pi 5 is an excellent solution for prototyping mobile robotics. However, the issues appear when you try to use power-banks and batteries in order to deploy Raspberry Pi 5 remotely on a mobile robot.

One of the main "issues" with Raspberry Pi 5 is that it is a power-hungry device which requires 5.1V and 5A. This is a non-standard power requirement since most standard USB-C type of adapters and power-banks provide 5V and 3A. You have to do all sorts of converter acrobatics to provide stable 5.1V and 5A, or you have to purchase a custom designed lithium-ion battery adapter. These custom adapters might be unstable, and you have to be very careful dealing with lithium-ion batteries.

Instead of doing all this, you can simply purchase a lightweight power bank with a standard AC outlet, and use the standard Raspberry Pi 5 adapter, and you are good to go. There are affordable 20,000-30,000 mAh - 80W power banks that weigh only 1-3 pounds on the market. Since they are lightweight, a smaller mobile robot can easily carry them, and in addition, you can use them to power-up a camera, lidar, etc.

Here is a YouTube tutorial on how to couple Raspberry Pi 5 with a power bank and how to test the power bank such that it meets the current and power requirements:

https://www.youtube.com/watch?v=N8kT0...

8 months ago | [YT] | 11

Aleksandar Haber PhD

Alibaba just published the QwQ-32B model which is probably the best 32B parameter model that can be executed on a local computer. The performance is comparable to much larger DeepSeek 671B model. Here is a tutorial on how to install and run QwQ-32B model locally

https://www.youtube.com/watch?v=FSqbZ...

8 months ago (edited) | [YT] | 5

Aleksandar Haber PhD

If you have a GPU with more 8GB of VRAM you can run this amazing text-to-video, image to video, and text to image model. The name of the model is Wan2.1. Here is a video tutorial on how to install and run this model locally. On my NVIDIA 3090 GPU takes around 7 minutes to generate 5-10 seconds of high-quality video material.
https://www.youtube.com/watch?v=4YGLx...

8 months ago | [YT] | 6

Aleksandar Haber PhD

Here is probably the only tutorial explaining how to correctly implement AI agents locally using the n8n framework, Ollama, Llama, and other components. The tutorial is almost 45 minutes long and thoroughly explains all the steps you need to perform in order to run AI agents on your local computer. You do not need to have an OpenAI API and you do not need to pay a single cent to run agents on your computer that will access the internet, retrieve information, and process information. Also, this approach does not require knowledge of Python which significantly increases the spectrum of people who can implement AI agents. However, the implementation procedure is still not straightforward since there are a number of different components that need to be interfaced. Enjoy following this tutorial and consider supporting this channel for more free video tutorials:
https://www.youtube.com/watch?v=6S49k...

8 months ago | [YT] | 11

Aleksandar Haber PhD

After more than 8 years of teaching and working with Robot Operating System 2 (ROS2), and after creating a number of popular tutorials and working with a number of students and engineers, we realized that the "best" and easiest entry point for learning ROS2 Jazzy might be the built-in Turtlesim simulation. This is a simplified 2D model of a mobile robot. However, the Turtlesim simulation has almost all the components that you need to implement in order to control a real robot using ROS2. By carefully studying this simulation and all the nodes/topics behind the ROS2 communication graph, you will get a solid understanding of how the ROS2 system works behind the scenes.

Here is a tutorial on how to start with the Turtlesim ROS2 simulation: https://www.youtube.com/watch?v=k3NHS...

8 months ago | [YT] | 7

Aleksandar Haber PhD

Upon the request of a large number of subs and supporters, here is a tutorial on how to install and run RAGFlow locally with local LLMs on a Windows computer. This tutorial enables you to run privately and securely your own Retrieval Augmented Generation (AI) AI system, and to organize all of your documents in a functional database that can be searched and analyzed by a local LLM and an AI agent. Companies are paying a lot of money for something that they could do 100 time cheaper by using RAGFlow (or some other RAG systems that will be covered on this channel):
https://www.youtube.com/watch?v=DLI65...

8 months ago | [YT] | 5

Aleksandar Haber PhD

Here is how to install and run locally a very powerful text-to-speech and voice cloning AI model called Zonos on a Windows computer. This model was originally developed for Linux Ubuntu, however, after a number of trials and errors I was able to run the model on Windows using WSL:

https://www.youtube.com/watch?v=oCWPh...

8 months ago | [YT] | 9