A series of tests made together with students for the purpose of Tangible Interaction Design classes at Industrial Design faculty. I decided to put it online as a reference for future projects.

I didn't put all sensors code because some libraries and code examples provided are straightforward

Sensors tested:

Ultrasonic sensors:

• HC-SR04: We had two kinds of these. We've tested various tricks to improve their measurements.
• DFRobot Gravity URM07
• DFRobot Gravity URM09
• DFRobot Gravity URM37 v5

Infrared sensors:

• Sharp GP2Y0A41SK0F
• Sharp GP2Y0A21YK0F

Time of Flight sensors:

• GY-VL53L0X
• Pololu VL53L3CX: There are no Arduino libraries or registers documentation available. Tested on STM32F446RE and STM32duino.

Gesture sensors:

• APDS-9960: Detects Left/Right/Forward/Back/Up/Down gestures.

Testing conditions:

Locations:

• Open space.
• Hallway wide and narrow.
• On a table.

We've tried to test with and without background elements.

Measurments:

We've measured different kind of objects with:

• Flat surface.*
• Uneven surface.*
• Narrow object.
• Material.
• Hand movement.
• Pedestrian.

*Small and larger object.

Review of results:

HC-SR04

Before using this sensor, lower your expectations, otherwise the disappointment will hit you hard. People who praise the quality of sensor use them in particular case measuring large and flat objects. Sensor provides a good result if it gets the object, which in some cases is difficult.

Maximum range:

• Some vendor's datasheets declares up to 4 meters range. Unfortunately there is no way to use it effectively for distance over approx 2.5 meters unless you measure a distance to a large flat surface. It's rarely the case during our classes.
• Distance to hand is up to 50cm.
• Measuring distance to a person walking toward your project is up to 200cm.
• Small object on a table is up to 1m.

How to fix the problem with hanging?

There are many variants of this sensor. However, chepest version has issues you can read about here: http://www.therandomlab.com/2015/05/repair-and-solve-faulty-hc-sr04.html to be honest using the cheapest no-name sensors without the proposed solutions is worthless.

• Use the included code according to mentioned article (it's in comment section!). I made some modifications to it.
• Solder the 10k resistor according to the picture. This prevents from hanging.
• Add 470-1000uf capacitor between 5V and GND. Unfortunately we don't see any difference.
• Quick and dirty solution I found personally: reset the sensor by switching it on and off via transistor. Use it only with long intervals of measurements, you probably decrease its lifetime by a significant value.

There is a New Ping library. Notice how single and multiple sensors are declared.

Notes:

• It seems that only the cheapest versions require the mentioned hacks.
• You should choose different sensor when trying to control something with hands or detect small, non-flat objects.
• I provided various averaging methods, it makes sense when high noise is expected. One of them uses linked list data structure in which you don't have to shift all the values each time. Without discarding incorrect zeroes there is no benefit of averaging. Implementation depends on your project so it's not the solution to fit all projects.
• Reading multiple sensors can be slow. Computing averages of results to reduce noise would be O(M * N) slower.
• When increasing maximum range, increase the delay between measurments to avoid random noise caused by timed-out measurements.
• In code without New Ping to change maximum distance change the number in pulseIn function.
• Hack with resistor makes the sensor work louder, you can hear the clicks.
• The sensor's accuracy can be improved with thermometer (speed of sound varies with air temperature https://www.omnicalculator.com/physics/speed-of-sound TODO:convert HC-04 formula to compensate temperature).
• The code without zeroes implements a linked list which distinguishes between incorrect measurements and the zero as an actual value (when value is zero more then 4 times in a row).

DFRobot Gravity URM07

Full guide: https://wiki.dfrobot.com/URM07-UART_Ultrasonic_Sensor_SKU__SEN0153

Code provided in the above link is made for Arduino Leonardo (Atmega 32u4) Serial and Serial1 ports. I' made a software serial example to make it work on Arduino UNO.

Maximum range:

20-750cm

Notes:

• URM07 has built-in temperature sensor.
• URM07 communicates via Serial.
• Detecting angle is 60 degrees, it's important if you intentionally want to detect surrounding things so the sensor is not suitable for narrow spaces. However, it looks like it's better than URM37.
• You can have many of sensors on single bus by changing their addresses.

DFRobot Gravity URM09

https://wiki.dfrobot.com/URM09_Ultrasonic_Sensor_(Gravity-I2C)_(V1.0)_SKU_SEN0304

Maximum range:

2-500cm There are different ranges 150, 300, 500. According to documentation lower range will reduce sensitivity?

Notes:

• URM09 communicates via i2c. It also has its analog equivalent available to purchase.
• URM09 has built-in temperature sensor.
• URM09 has very large measurement angle, but not as high as other DFRobot mentioned here.

DFRobot Gravity URM37 v5

https://wiki.dfrobot.com/URM37_V5.0_Ultrasonic_Sensor_SKU_SEN0001_

Maximum range:

2-800cm

Notes:

• URM37 communicates via Serial but it has various operation modes. It has additional features comparing it to other DFRobot sensors mentioned above.
• URM37 has built-in temperature sensor.
• URM37 has a button to switch from Serial (TTL) to RS232 communication. Do not play with that!
• URM37 has very large measurement angle, it grabs everything and can't work correctly inside narrow hallway. It's suitable for large scale measurements.

Sharp GP2Y0A41SK0F and Sharp GP2Y0A21YK0F

Sensors are similar.

Maximum range:

• GP2Y0A41SK0F: 4-30cm
• GP2Y0A21YK0F: 10-80cm

For both you can use SharpIR Arduino library, if you check the library .cpp file you will see how simple is the measurement equation.

Notes:

• Analog sensors win in terms of simplicity of connection and they're very accurate.
• They have a large blind spot (13.3% / 12.5% of their total range) which must be taken into account.
• Theoretically you can compute polynomial regression model to calibrate your sensor to improve accuracy. I guess my students aren't proficient in machine learning nor statistics so I won't explain it further.
• Sunlight can cause noise in measurements.

VL53L0X

TODO: Figure out how to calibrate the sensor to check the difference. Without calibration sensor is fine.

Maximum range:

• 120cm in standard and high accuracy mode. (approx. 1m effective)
• 200cm in long range mode.
• Hand movement: 80cm in long range, 50 in high accuracy or default.
• Very thin object up to 30-35 cm.

Notes:

• They're relatively cheap comparing to what they offer if you buy a no-name board rather than Adafruit.
• The cheap board (violet color) is the smallest among all which matters in some situations.
• You can use Adafruit library.
• Straightforward usage is a plus. You can choose between long range and high accuracy modes.
• Long range or high speed mode is not suitable for outdoor daylight. Outdoor usage cuts the accuracy and distance dramatically (Datasheet 5.3 Ranging Accuracy). Direct exposure to light can create noise in measurements.
• High accuracy mode is slow and even in that mode the value floats +/- 3mm so eventually the fact that sensor returns distance in milimeters won't give you that much. Maybe computing averages in faster modes make sense? Measurement frequency is faster than ultrasonic sensor.
• It is possible to calibrate the sensor but guidance is not provided through Adafruit library.

Pololu VL53L3CX

Because of the way it works, the sensor can't be used with Arduino boards at all. I've tested it on my STM32 Nucleo F446RE with ARM Cortex M4 with STM32Duino library, but everything mentioned is out of scope for our classes. What is more - I don't feel competent enough to do reverse-engineering of ST propertiary driver from scratch.

VL53LCX measures distance through computing histogram making it capable of capturing multiple objects within its field of view and returning their distances separately. Unfortunately there must be a significant difference between them.

• Maximum range: 310cm indoor / 100cm outdoor (Datasheet 5.3) measured in scale of milimeters. The only interference is the sunlight which dramatically drops the max distance and accuracy.
• Detects thin object and works fine with soft materials, but requires calibration offset (I measured the distance with ruler and it always had approx. 3cm too much no matter what kind of object was measured).

Tips on STM32duino and VL53L3CX:

• Here is the guide to install https://github.com/stm32duino/wiki/wiki/Getting-Started you also need to install STM32CubeProgrammer and include it into system's PATH
• You can download STM32Duino VL53L3CX library in your library manager.
• VL53L3CX XSHUT pin is XSDN_I according to example sketch.
• Actually the address 0x12 is assigned on runtime. Default i2c address is 0x52. You don't need to change it.
• VL53L3CX pins aren't 5V compatible. Pololu board has voltage regulator but they wrote that XSHUT and GPIO1 aren't level-shifted.
• TODO: add calibration code.

APDS-9960

Measurement distance is up to 100mm (in my opinion it's up to 150mm), however the measurements are raw IR light density values not distance units.

It can measure up/down/left/right gestures and near/far when using proximity data.

Notes:

• Sensor operates in 3.3V logic, do not connect VCC to 5V. SDA and SCL will work on 5V however it's not recommended (you can use logic-level converter).
• There are libraries from Adafruit, Sparkfun and Arduino. Their example code looks similar. Some of them use interrupts so you can respond immediately whenever an event occurs instead of polling for it. Sparkfun library includes near/far gesture based on hand distance but Arduino library looks like is not blocking the main loop.
• Within the libraries it seems that measuring proximity and gestures doesn't play well with each other.
• PAJ7620U2 seems to be an alternative, it computes the gesture internally?