IPS PX4 Bootcamp · No GPS signal in your lab?
Indoor Positioning Lab with Pixhawk
A 6-week live online bootcamp, Mondays 4 to 6 PM Pacific Time, for UAV and UGV engineers who have to navigate where GPS cannot reach, without the 2-year master's, the $50K motion capture rig, or the years of trial and error.
6-week workshop
I only accept 5 seats at most.
Live online, $1,800.
Apply by Sep 19. Starts Sep 21.
You build the infrastructure underneath every robot: indoor positioning at 90 Hz, from ceiling cameras into a Pixhawk running PX4, and the PX4 architecture and ROS2 communication that carry it into the Pixhawk. It is not a software-only course. You put up the poles, mount the cameras, print and glue the camera calibration targets, and pipe the real-time position into the Pixhawk (+ Jetson, optionally) yourself.
Who is hosting this?
Elliot, an aerospace GNC (Guidance, Navigation, and Control) engineer who builds UAV and UGV systems for a living. About 190 students and working engineers have learned from his recorded GNC lectures, and some of them have since been hired as GNC engineers. This bootcamp goes where online lectures cannot: a hands-on workflow, built in your own room, with him live online every week.
This is NOT for the hobbyist who wants a cool robot by Christmas.
Why your robot gets lost
Six messages I received
Someone wrote to me a while ago. The robot is built, and the mechanics and wiring are done, with a Pixhawk on board. What he does not have is a robust position sensor for the robot indoors. He wanted, in his words, “positioning like in a simulation, like Isaac Sim.” A fixed frame he can trust.
Another client is building a UAV with PX4. His question was about the velocity estimate, where a reliable velocity source comes from when GPS alone is not enough.
Someone else told me he wants to avoid camera calibration altogether, because it is not trivial. He is right. It is not.
Another person was stuck on his drone's camera parametrization for autonomous navigation.
And one more asked me whether I feed optical flow as a velocity sensor into the Pixhawk.
And finally, one more said, in his words, “I don't even know how the PX4 software architecture works, let alone how to feed all this into it to complete the position loop.”
Six people, six robots, the same bottleneck. None of them can get the real-time position into the vehicle so that it follows their path and waypoints in a GPS-denied situation. A robot that does not know where it is cannot navigate, and a lab that cannot measure where the robot is cannot debug it.
Is this you?
My algorithm only lives in simulation. At the very least, I want to test it hardware-in-the-loop (HIL).
I want to implement the latest VLA papers, but I can't test them because GPS doesn't work in my building.
I want a $50,000 Vicon motion capture system, but I don't have the funds yet.
I have a robot, but I can't debug it to navigate autonomously indoors.
I need a higher update rate than UWB or ultrasonic sensors, like 60 to 90 Hz.
I need my own R&D process for unmanned vehicles with Pixhawk/PX4, one that is robust, consistent, repeatable, and scalable.
I need a more interactive workshop than recorded lectures to set up an unmanned vehicle lab.
I use ROS2, but I have never opened the PX4 source, and I do not know how my messages actually reach the vehicle.
I have optical flow or SLAM on board, but nothing to check it against, so I cannot tell how far it drifts.
Are you stuck piecing together a robot R&D workflow by yourself?
You have the programming skills and the basic math, but the workflow is still all over the place?
I was stuck too
All of these were my stories too.
Some years ago (before the ChatGPT-3 era), I dropped out of my PhD in the very first semester, right after my MSc, rented a warehouse, and started building my own UAV and UGV systems. I imagined myself becoming one of those entrepreneurs who drop out of school and build the big things.
Then I hit the wall. The big dogs had something I did not, and it was not the robots. It was a room that always knew exactly where the vehicle was, a $50,000 Vicon system on the ceiling. Every test gave them a grounded number while I was eyeballing. I could not move forward like that. I could not find $50,000 all of a sudden. I was frustrated, and for a while I gave up on the idea.
But the voice in my head would not let it go. I am an aerospace GNC (Guidance, Navigation, and Control) engineer with years of research behind me. More importantly, I had a customer waiting for a tailsitter autopilot, my first real contract, the first time my own robots would pay me back. I could not walk away from that.
So I built the indoor positioning system for Pixhawk/PX4 myself, on my own budget. Two things, and both have to work. A camera system in the room that tells you where the vehicle is, consistently and in real time. And a Pixhawk that takes that position and runs your code, so the vehicle actually goes where you tell it.
After some desperate attempts, it finally worked. The Pixhawk's position followed my vehicle across the floor without a jump, a smooth, continuous position signal. From then on, every test gave me a 3D position vector instead of a guess or jumps. For the first time since I started my business, I felt in control of my own work. That was the day I became a 1-person R&D department, one that could ship a product by itself.
A few cameras running at 90 Hz on the ceiling, $2,500 instead of a $50,000 Vicon system. If someone handed me $50,000 today, I would not spend it on a Vicon. My 90 Hz is enough. And the lab turned out to be more than a position sensor. It is the bridge between simulation and the real vehicle, sim-to-real and real-to-sim, the same infrastructure you need for a digital twin or for testing VLA models on real hardware. I have shipped a tailsitter UAV autopilot product to that customer with exactly this setup.
In 6 weeks, I will help you build the same real-time pipe to your Pixhawk in your own garage/room.
What it costs to get here
- The big dogsVicon motion capture rig
- $50,000
- Graduate schoolA 2-year master's, tuition alone
- $40,000
- This bootcamp6 weeks, live online, your own lab at 90 Hz
- $1,800
Plus the cameras and Pixhawk you buy yourself, $600 to $2,500. See section 06.
Two things, both have to work
- 01A camera system in the room that reports where the vehicle is, in real time.
- 02A Pixhawk that takes that position and runs your code.
What it turned into
A 1-person R&D department, a shipped tailsitter autopilot, and the bridge between simulation and the real vehicle.
What you can achieve
Today
After week 6
Why not a Holybro H-Flow or a vendor's SLAM instead?
Remember the control group and the experimental group from elementary school science class?
One position source is an experiment without a control group. The cameras on the ceiling are your control group, so an optical flow sensor or a SLAM stack on the vehicle gets measured against them instead of trusted. That is a proper experiment, and it is yours. No vendor lock-in, no waiting for the vendor's next update.
Because you can change one gain, run again, and read the result a few seconds later, you feel in control of your robot for the first time. That is the debugging loop, the R&D loop. Idea, test, number, next idea. One iteration takes an evening instead of a week. A controller that used to take you weeks, for example, is tuned in days. You are set free from the endless simulation world. And you are finally working in the real world, as a roboticist.
From that day, you are not a person with a robot anymore. You are a 1-person R&D department with a lab.
By the end of 6 weeks, you will have
The hand-held test, passed
The BOM
Camera calibration workflow
Reusable visual markers
Your own indoor positioning system
The full position pipeline
A working map of PX4
The math behind every step
A well-funded university lab has this with a $50,000 motion capture rig and a team. After 6 weeks of the bootcamp, you have it with a few cameras, a Pixhawk, and yourself.
The benefit compounds. Every vehicle you build after this starts with positioning already solved, so the next project skips the weeks that stall most robot projects and goes straight to the algorithm.
How the 6 weeks work
There are two usual ways to get here. One is a master's program, a minimum of two years, a $40,000 tuition bill, and a lot of coursework you will never touch again to get to the few things you actually need for a vehicle that flies or drives. The other is trial and error on your own, burning money and months with every failure. I went through both of these.
This bootcamp takes a detour around both and goes straight to the point. Six weeks of live online sessions, Mondays from 4 to 6 PM Pacific Time, for working professionals. You build a scalable R&D platform yourself, alongside your job, without the $40,000 debt and without the wasted years.
6 live sessions of about 2 hours, one a week, Mondays from Sep 21 to Oct 26, 4 to 6 PM Pacific Time (US West Coast, UTC-7).
Weeks 1 to 3: cameras, calibration, markers, and the positioning system. Weeks 4 to 6: the PX4 architecture, its apps, modules, and uORB data flow, MAVLink and uXRCE-DDS to ROS2, and the hand-held test, your position live inside PX4.
Optional bonus at the end: if your vehicle is ready, we put the Pixhawk on it and demo waypoint navigation. This is a demo, not part of the guarantee.
A take-home project every week. Vibe coding is fine where it applies.
You read and present research papers. I do not just give demos and lectures.
I answer the chat at least once a day.
Every session is recorded, and the recording is there for you if you cannot join live.
The exact order of the curriculum may shift to fit the cohort's pace and hardware arrivals. The scope of the 6 weeks does not.
Weeks 1 to 3 build the indoor positioning system. Weeks 4 to 6 open up PX4 and feed that position into the Pixhawk. One live session a week, a take-home project every week, and member paper presentations along the way.
Week 1
Onboarding
Goal: everyone runs the same environment and can pass messages between machines.
- Docker setup
- ROS2 / Zenoh / Iceoryx2 with a virtual robot
Week 2
Camera calibration
Goal: turn a raw camera into a calibrated sensor and understand the math behind it.
- Calibration walk-through
- IPS calibration in simulation
Week 3
Your positioning system
Goal: cameras are on the ceiling and your room publishes a position at 90 Hz.
- Multi-camera setup
- Broadcasting the position to a phone or the network
Week 4
Pixhawk and PX4 architecture
Goal: PX4 stops being a black box.
- Modules and uORB, writing a small PX4 app
- Driving motors from Python
Week 5
Position into the Pixhawk
Goal: the IPS position reaches PX4 and the hand-held test passes.
- MAVLink and uXRCE-DDS
- System identification in simulation
Week 6
Your project
Goal: every member presents a small project built on their own pipeline.
- Checking optical flow against the cameras
- A waypoint demo on an RC car, if the vehicle is ready
What you bring
Hardware budget
$600 – $2,500
You prepare or buy the hardware yourself, from the vendors. I do not sell or ship any parts. The full BOM, with part links, comes on day 1, and you need everything in hand by week 3. If you are starting from zero, with no Pixhawk and no machine vision cameras, budget about $600 on top of the workshop fee for the budget path, up to $2,500 if you go for the full high-end gear.
Machine
Ubuntu 24
You also need a PC or laptop running Ubuntu 24 (no VM, no WSL).
Time
10 – 15 h / week
Plan for about 10 to 15 hours a week. The session is 2 of those hours, and the rest goes to the take-home project and the paper you present.
This is not a software workshop. This is robotics, and your body moves, not just your fingers. The cameras go up on poles that stand on the floor and reach toward the ceiling, so you can run the whole lab in a rented room without touching the walls. If you own the space and prefer a permanent install, you can bolt them to the ceiling instead. Either way, you run cables, print and glue calibration targets, measure them with a caliper, wire a Pixhawk, and walk the marker cube around the room while the cameras watch it. Plan for some open floor space.
One expectation to set on the result. This lab is accurate to a few inches. I am not saying this is an industry-grade, millimeter resolution system. However, that is not needed for your first product, MVP, or proof of concept. What you want is a robust indoor positioning system that your robot can simply subscribe to, like GPS, without the hassle.
Reconsider if
The program can be intense. Please reconsider if any of these describe you.
- You CANNOT free up the time for 6 weeks. It takes about 10 to 15 hours a week.
- You just want a working code base that you do not understand, from your desk. This one keeps you on your feet.
- You HATE reading research papers, and you never want to think in statistics or about software-to-hardware architecture.
- You want to remain a ROS2 package consumer forever, with a blind belief in “never reinvent the wheel,” even when the framework dictates your architecture.
- You have NEVER written a line of code or connected battery wires to a hobby robot (or Arduino).
Made for you if
This will be a comfortable workshop if these describe you.
- Your work already involves R&D and research paper reviews, and you are fine presenting one to the group.
- You have written code and wired a hobby robot before, in any language, at any level. You do not need to be an expert at either.
- You use ROS2, PX4, or a similar stack, and you want to know what happens under the hood instead of trusting the package.
- You enjoy deep thinking: deriving the math, reading a data sheet, and testing an idea against a number.
- You are independent and self-directed. You can put up a camera pole, follow a BOM, and finish a take-home project without being chased.
My promise to you
The guarantee, in one sentence: if you meet all four conditions below and, at the end of week 6, your Pixhawk still does not subscribe to the indoor positioning system (IPS) position from tracking a fiducial marker in real time, you choose between 2 more weeks of 1-on-1 support from me at no charge or a 100% refund of the workshop fee.
The four conditions. All four must be true.
01
02
03
04
The test is simple. You carry the marker cube with the Pixhawk across your room, and the Pixhawk's linear x, y, z position estimate (not the angular pose) follows you in real time. Push your repo and the Pixhawk log from that walk by the deadline, and we both have the proof.
No hardware or no commits, no guarantee. I cannot guarantee what you did not build.
The guarantee covers the workshop hand-held test. I do not consult on personal projects.
If the hand-held walk still fails at the end of week 6, we start with one diagnostic call, where you run the failing test for me on your hardware and walk me through your repo. Then you pick: up to 2 weeks of 1-on-1 support after the last session, at no charge, or a 100% refund of the workshop fee, requested within 7 days of the last session. For the 1-on-1 sessions, you bring the data and the code.
Only 5 seats
Price
$1,800
Seats
5
Applications close
Oct 3
Starts
Oct 12
The bootcamp is $1,800. This is the first cohort, at most 5 seats, applications close Oct 3, and we start Oct 12. It is small, and it leans on me being in the chat and on calls with you more than a larger group would allow. If you ever want to break out of a muddy workflow at once, now is the best time, since this is my new training program and you can interact with me closely. That is what the price reflects, and it will not hold as the cohort iterates.
Apply now
If this sounds like your situation, fill in the application below. I read every one myself, and the system will automatically guide if this workshop fits you or not. If any notes, I will read and let you know.
Elliot
Claim a seat
Join IPS PX4 Bootcamp
Checking seat availability…
- Price
- $1,800 (tax included)
- Seats
- 5 seats
- Dates
- Oct 12, 2026
- Timezone
- America/Los_Angeles