Lesson 3

Data Logger

Data was logged from the microphone and plotted in a chart using Excel:

Chart showing noise

Sound Controlled Car

The car moves based on sound meaning that when a noise is made the car switches between the following driving modes:

• Forward
• Right
• Left
• Stop

This results in a car moving a bit odd. The way the program makes sure that random background noise doesn't result in change of direction is by having a minimum thresshold. Only if the dB level of the sound is above 90 a change will occur. 

Clap Controlled Car

In the last part of the Lesson, we experimented with implementing the SIvan Toledo algorithm for detecting a "clap". He described a clap as a sequence of key characteristics that must be met within a certain timeframe of each other. Specifically a clap is instantiated with a "low-amplitude" sample of below 50, followed by a very-high amplitude sample of at least 85 within 25 ms of the instantiation and followed by a low-sample below 50 within another 250 ms of the high amplitude. In order to filter out noise from potential instantiations of a clap, we used the data collected from the second part to determine a noise level threshold, which we determined to be 30 or below. Using this knowledge, we ignore all samples below this value when listening for a clap instantiation, i.e. to instantiate a clap we listen for a sample between ]30,50]. Because this algorithm is very time sensitive, we maxed out the sampling rate at 3ms interval sampling. One thing we experienced with this alternate way of detecting a clap, is that it is a bit more sensitive to the actual sound of the clap, we needed to clap in a certain way for it to detect it as a clap. It actually seemed more sensitive to snapping your fingers than the common clap, we experienced that our claps needed to be very high pitched, short and loud. We expect that we would have been able to work out the quirks and inaccuracies given more time to tweak the individual thresholds, timeouts etc. But all in all, it seems to be more picky about the exact sound of the clap, but also more resistant against other loud noises.

Lesson 2

For our second lesson we were to learn how to use the ultrasonic sensor for the NXT.

To that end the first part consisted of uploading a testing program for the sensor, which verified that the sensor was actually operational.

Tracker

Thereafter we uploaded a Tracker program demonstrating the use of said sensor. It is a reactive program, that causes the car to approach an object in range of the ultrasonic sensor until the distance to it is reduced to 35cm. Seeing no object in range it will drive forwards. If the object is closer than 35cm the car will back up. The car will however not at any point stand still.

Wall follower

The core of our second lesson was to program the car to follow a wall. We achieved this by using the tracker program as a basis. We ended up trying two different methods of implementation.

Both implementations work on the principle of letting the car constantly move forward while adjusting the power to the left and right wheels, depending on the correction (if any) to be made.

The first implementation involved adding a threshold (error margin) constant which let the car travel straight for a while before making a correction to the error that has been accumulating by not travelling parallel to the wall.

We did this after learning from lesson 1 and the results of other students efforts during the lesson, that correcting an error instantly results in a jagged and inefficient movement pattern.

Besides this and tweaking (mostly via trial and error) some other constants like the desired distance to the wall and the extent of the force with which an error was corrected, no other significant changes were made to the tracker program.

We were pleased to see that this implementation actually worked quite well and moved very efficiently along the wall.

In an effort to make the wall following routine more concise and allowing the car to correct different magnitudes of errors with an appropriate response, we implemented another solution based on a different principle.

This time the car would constantly correct its distance to the wall by adjusting the power to the wheels based on the square root of the error from the desired distance to the wall. After some erratic results and obtaining the insight that a form of threshold was still necessary we managed to make this solution work as well.

The primary objective behind our second attempt was to try a different and more implicit approach to allow the car to turn around corners, since implementing that behavior in the former approach would mean a lot of tweaking or the implementation of a new subroutine just for that purpose.

Unfortunately we did not manage to make either of these programs capable of turning around neither inside nor outside corners.

We found that our first solution worked better than the second one, although we suspect that we might have chosen the wrong function for correcting the error in the second one.

Comparison of NCQ implementation

Philippe Hurbains NCQ implementation of the wall follower uses the three state algorithm, which turns hard (read: in place), when it is too close to the wall and shallow when only a minor correction is needed.

Lesson 1

Write text here