Determining sound direction

I've been playing a bit more with datalogging (check out the fantastic tips from Brian in the forums) and I've always wanted a robot that could come to me when I called it. The sound sensor is great for 'detecting' sound but no good for deciding where the sound came from. One way to figure out where the sound is coming from is to use 2 sound sensors.

So I built myself a little NXT head that had 2 sound sensors and a bit of foam in between.

"Sure it's pretty" I hear you say, but whats that foam in between the ears for?

Well I figure that if I had 2 ears, with some sort of dampening material in between, the ears would hear the same sound source, but at different volume levels.

ie. If I clapped my hands very loudly on the left hand side, then the left 'ear' would hear a certain volume and the right 'ear' would also register a sound, but at a much quieter level.

So I set up my head and start writing sound readings from both the left and right ears to a text file as fast as I can. I'm not too concerned with the sampling rate, as I'm more interested in the difference in sound levels for the same 'clap'

I take the results and put them through Excel where I subtract the value of the right ear from the value of the left ear to give me a difference.

Now if the clap happens directly in front (or behind) my NXT head, then the sound levels will be the same and the difference will be 0. But if I clap off to the left hand side, then the left reading will be higher than the right reading and my difference will be a positive number. If I clap on the right hand side then I'll get a negative number.

This graph shows the difference between the left and right ears with 3 test claps, once on the left, once in the center and once on the right.

From this I can see anything under -40 is a definite clap on the left and anything over +40 is a clap on the right.

So now I can write a program that takes the two sound sensor readings and subtracts one from the other. If the result is more negative than -40, I turn clockwise a small amount. If it is greater than 40, I mover counter-clockwise for a small amount.

I think with more sound dampening, I could probably get it to respond a little more accurately.

Damien Kee


Joshua Heinzl said…
This is similar to what one of the groups during the Engineering Symposium did. They had 2 light sensors separated and measured the difference, which would be input as the motor direction and power. The goal was to follow a light.

If where the sound is coming from is constant, you could rotate the sensor and stop at fixed points, then see when the value was highest, and set your heading that way. Of course with 2 sensors it is much faster by comparing the difference.

Damien Kee said…
Sounds like they were making Braitenberg vehicles.

I would have loved to have gone to the symposium but it was on the other side of the world.

Anonymous said…
I tried this on my robot the 'nsb Bandit' (last seen on the NXTLOG sumo challenge)

I never tried data logging but my results were pretty good. I always had the robot looking for sounds, but it had to be timed just right to turn the proper direction in practice. It worked most of the time, great as a proof of concept.


The program is in the nxtlog
Anonymous said…
This bot is very good!

I also made a NXT robot with 2 sound sensors.

Please refer my video.
Thank you!
Unknown said…
Sound detection and procession is a intriguing field for robots indeed where bioengineering patterns play a prominent role, in particular from the insects' universe.
I recently read some article about the detection of sounds and their direction by male crickets - their sound "sensors" are placed on both sides of the body in such a way that the signal of the sound sensor on the far side of the sound source is terminated by superposition from the near side's sensor signal when the sound's frequency is the same as the one that female crickets use. Fascinating!
Anonymous said…
Have you tried three sound sensors? with maybe a centre detecting an overall level of sound? at present a not so loud sound might not register.
Alternativly you could use left to centre versus right to centre?
Anonymous said…
I too have made a robot that moves to sounds.

I used pieces of an egg carton to focus the sound better.

I wanted 2 nxts to track each other with timed beeps, but the sounds were not loud enough.

AlexD said…
Yeah, I second the idea of "ears". Most larger animals have very specialized pinnae in addition to the acoustic shadow provided by the head. So, egg cartons, funnels, paper plates, all have helped when people in my class have tried to get some directionality. And yeah, Braitenberg rules! You should see how two lines of code made a robot avoid walls as an artefact of being attracted differentially to light! Of course, knowing some control theory does not hurt either :-)
Unknown said…
Any idea where your initial transients are coming from?

At first, I thought it might actually be from the delay in arrival between the two sensors. There's no horizontal scale on your graph, but sound would take about 0.6 ms to travel 8 inches at sea level, and from a guesstimate of the length of a clap, it might be in the ballpark.

But since the sign of the transient is constant for left, right, and center claps, that doesn't seem like a reasonable explanation. So is it coming from the brick sampling the left and right sensors alternately?

Popular Posts