vSlam with Ardupilot


I want to build a drone that will perform certain tasks using vSlam together with Ardupilot. I have been doing research on this subject for almost a week and I am trying to learn more about vSlam by reading the writings of developers who have realized such projects before me.

Especially these 2 previously developed projects helped me a lot:

  1. Experiment with Visual Odometry - ROVIO

  2. https://discuss.ardupilot.org/t/integration-of-ardupilot-and-vio-tracking-camera-part-1-getting-started-with-the-intel-realsense-t265-on-rasberry- pi-3b/43162

Thank you @ppoirier and @LuckyBird for these.

If we go back to my project, the task I want to do is to realize a drone project that will read the QR codes or barcodes of the products in a large warehouse. Before moving on to this mission, I first need to build a vehicle that can fly indoors. I’ve looked at the 2 examples I’ve shown above and many more, but I couldn’t find a complete method because I’m inexperienced with this vSlam.

Although the project developed by @LuckyBird was very suitable for me, unfortunately, the fact that the T265 cameras were discontinued as of February 2022 prevented me.

In addition to these, I have also reviewed each example found at this link:

I would be very happy if I could find an example that has been done or any guidance that can guide me.

I am open to your suggestions. Thanks in advance.

Warehouse drones is the Holy Grail of GPS denied configuration.
Unfortunately this is not an easy task as the warehouse environment is extremely challenging based on the size, compexity and the extreme repetability of the features that requires a lot of intelligence in the SLAM pipeline if you dont want to add pre-installed fixed markers to hep for navigation.

First of all, if I can develop a vehicle that can fly indoors with vSlam, I can progress step by step, perhaps working for a very long time, to develop something. I am currently researching all the vSlam techniques I can use. These include: ORB Slam 2 or 3, Google Cartographer, RTabMap techniques.
The biggest problem I have is a point how I can integrate them with an open source drone software.
My first goal is to develop a vehicle that can fly indoors with vSlam.

There is a similar project in the link that I want to do, maybe it may be of interest to you.

Link: Vision based Search & Rescue Micro Aerial Vehicles (MAVs) for catastrophic situations. Report for OpenCV AI competition on Vimeo

OrbSlam3 is a good contender, I could get statisfactory result using it with a Realsense D455 running on a I7 based UAV.

You can join the Discord Vision channel for specific questions about Vision Based Navigation

Thank you very much.

Semantic Slam is really Interesting, I tried Kimera once but it was quite complicated to get it to work

As for Avoidance with OAKD, I think @rishabsingh3003 could be a good ressource

1 Like

It’s an interesting technique. I’ve already started looking, thanks.

I reviewed the VOXL camera of the Model AI company from Ardupilot No-GPS techniques again. Seems like a possible solution for someone like me who is inexperienced with slam. Do you think I can come to a conclusion if I use it in my project? As an experienced persons, your advices is valuable to me. @ppoirier and @rmackay9

VOXL is a good option if you are looking for a lightweight indoor system. My experience with the VOXL CAM usage outdoor was not positive as the system can easily crash inflight due to vibration, even on a well turned quadcopter.

For the warehouse project you need to add a good mapping system that can cover a large area. Still, the size and homogeneity of the features makes it difficult if you cannot add complementary localization features like fiducial tags.

1 Like

Hello @davud_suleymanov
Did you manage to find a way to go for your project ?

Hello @davud_suleymanov. Luckily, I just started my thesis for my BsC in robotics with this topic. After all this months, my research is still unsuccessful, since there is no connection from a controller with vSLAM. So far both Ardupilot and PX4 works for ROS2, but I can’t take a final decision. Do you choose any of these autopilot?