Indoor navigation in 2021 - feasibility opinions?

Hi guys

First off i have searched and read all the posts i could about indoor navigation - but some are older, some say stuff cant be done that i then read is possible, etc, so - im hoping for some updated 2021 help here!

I have been proposed a project; I wont be coding it, but overseeing it.

However, i am in doubt if this is at all possible/feasible.

So here is the project - and im assuming GPS is not a possibility inside:

Step 1: Make a drone hover inside. a room, keeping its place - how much will this drift, whats the obstacles in this, and will this even work currently.

Step 2: Make a drone go from the hover point and say 1 meter forward, and hold its altitude. How much will it drift here - how precise is this?

Step 3: somehow ( to be determined ) feed the drone 3 points - a start a middle and an end - so for instance, go from one living room, to a doorway and then through the doorway. Feasible?

Now, i realise that the above might be hard - and might be expensive, but… is it doable - and if so, whats the prefered method/approach for this?

Any insight would be greatly valued!

1 Like

Step 1: It will work now in 2021 using optical flow or other methods.

Step 2: It will work now tin 2021 using optical flow or other methods.

Step 3: A LOT more complicated. We are working on this (mostly @rmackay9) Take a look at his github pull requests.

And here are the challenges:

  1. https://github.com/ArduPilot/ardupilot/issues/13903
  2. https://github.com/ArduPilot/ardupilot/issues/11026
  3. https://github.com/ArduPilot/ardupilot/issues/7192

I could imagine that a lot of the work for 3 could be done with ROS Slam packages but I’m not familiar with something like this, this is just the way I would approach this problem.

Hi - thank you so much for chiming in.

I will try to reach out to rmackay9 and see if i can get a bit of information regarding where the issues are.

Hi Vabe – thanks for your post.

I havent dived deeply into this, but i am about to : The ROS slam packages is a set of packages that would be helpful in doing what we are trying to achieve; but if say we wanted this to be done differently, could a c++ coder use those packages and somehow re-code/optimize them - or is this the wrong way of looking at it? Trying to get an overall sense of who to get, and how to approach it - if that makes sense.

Have you looked at this ?
https://ardupilot.org/copter/docs/common-non-gps-navigation-landing-page.html

What you described can be done with a Realsense T265.
Additionally, and you can use the D435 for for slam or a more lightweight avoidance has demonstrated in wiki.

Hi - thank you, this looks amazing - but its also pretty big for the purpose we are going for. But right now im soaking all the information, so very much appreciated!

Looks like @amilcarlucas and @ppoirier have given you a few links already. I’d just like to reiterate that I think this is already possible using the Intel T265 camera or possibly an optical flow sensor + downward facing lidar.

The requirement to go through the doorway is the most difficult if you actually want to avoid hitting the door frame. That requires a depth camera and bendy ruler path planning. Once you want this then you need another companion computer and camera and the drone starts getting large. I think it’s only a matter of time though before we will be able to do it with a single camera and perhaps no companion computer.