Guidance for autonomous drones?

[InterCity]

Hello, guys.

I'm thinking about modifying my quadcopter with Raspberry Pi, voice command and make it for example follow me. The problem is, I can't find a reliable way of navigating the drone towards myself.

Computer vision/blob detection? Get ready for advanced algorithms, blob detection won't do. Human being is too complex for that.

GPS? Not enough precision, need to be precise on the scale of centimetres.

ADF radio beacons/directional antennae? Don't know how this would work at such small scale.

Voice detection? Might just work, but unreliable due to ambient noises.

I'm basically asking, is there a reliable way of directing a small machine autonomously? Because it might be a funny gift for someone, and a good DIY project...

[Technical Ben]

IR dot/s. It can follow them easily with a simple camera (drones are now for sale locally with cameras in built, a filter over the top possibly?).

Blob detection of your t-shirt with a qr code/specific blob on it.

[InterCity]

I might just do the IR beacon thingy. Thanks a lot

[Ralathon]

IR would probably be best, as Technical Ben said. Just glue some IR LEDs to your T-shirt. From the distance between the LEDs you can figure out the distance and angle to the shirt.

There are a lot of parallels with the freetrack project. Freetrack is for head tracking, but the code is open source and you could probably copy a lot of it for your purpose.

[Tex_NL]

You might want to augment the IR tracking with GPS and/or a radio beacon. If for whatever reason your drone ever looses contact with the IR tag it at least knows in what general direction to search to reacquire it.

[Idobox]

You can use RF two ways to do that.

The cheap way is to use measure power:

Put an emitter on the person, for example a phone with bluetooth, put a receptor on the drone, measure power, which is proportional to 1/r², get distance. With off the shelf components, the signal might too powerful at this kind of range to be measured accurately, and require an attenuator (can be done with resistors). With two or more receptors, you'll know in what direction to move, with one, you will to "wiggle" to find out.

A variant is to use a RFID reader and passive tags, and measure RSSI (return signal strength intensity). You would have to use UHF, 2.4GHz or higher, because lower frequencies use magnetic coupling, which for a number of reasons, would be a nightmare to use on a drone (think contactless cards or NFC phones). The advantage over the previous is that passive tags are pretty much just stickers, which might be handier than a active radio source. The issue is that it would be more expensive, and not all RFID readers have RSSI.

The second option, which is a bit more complex, is to measure phase, ie time of flight.

To do that, the easiest way is to put an emitter on the drone (let's say 13.56MHz, free frequency) and a frequency doubler on the target (can be as simple as a diode with an antenna). You then measure the phase difference (can be done with a XOR gate) between the 27.12MHz signal that rebounds from the target to one you generate on board of the drone (using a diode again). At 27.12MHz, the wavelength is about 10m, so if you get a phase difference of 2*pi/10, it means the signal traveled 10m/10 = 1m, and since the signal traveled form the drone to the target and then back, the distance is 0.5m.

The first solution is simpler, but a lot less accurate. Antennas don't transmit with the same strength in all directions, so if you tilt the drone or the beacon, the system will detect a change in distance even if it doesn't happen. Also, as the components heat up, the calibration will change a little bit, and cause a small error.

The second solution is a lot more accurate, but a little bit more technical. Calibration isn't much of an issue: you are comparing two signals coming from the same source, so you only need a stable frequency source, which costs a few $.

[InterCity]

Thanks for all the ideas, guys. Here's how I solved the issue: I've put six equally IR diodes on the rear of my collar, and programmed the drone to hover in the air 1.8 m high (Downward pointing ultrasonic range finder for $8 on dealextreme). The drone is then programmed to follow the strongest source of IR light in sight, and maintain distance of about 1,5 metres from me (computed via the visible spaces between IR diodes and assured by another, this time forward-facing ultrasonic range finder). If for any reason the drone loses contact with IR source for more than three seconds, it switches to bluetooth guidance and will try to follow my mobile phone while avoiding any obstacles. I didn't test it in real world yet, but at least for my laptop that I used to simulate the thought process of the future drone, it seems alright.

[spudman2]

Let us know how that goes, although I hope the drone doesn't see the sun or a car as the "strongest IR" nearby. Poor driver is suddenly pursued by a drone...

[Technical Ben]

So a little personal flying assistant? I could use one of those!