Low Cost 9-DOF IMU

[ynsta]

This one seams pretty interesting :

http://store.diydrones.com/ArduIMU_V3_p ... imu-30.htm

Someone tried it ?

[cybereality]

Haven't tried it, but they have some cool gear at that store.

[Krenzo]

Yes, I have this IMU. I had a lot of trouble getting this working and eventually ported another IMU library to it instead of using the software that is designed for it.

[virror]

Krenzo: Did you test with the latest v3 of the board and what kind of problems did you have?
Thinking about buying this one since it has a very good price.

[Krenzo]

Krenzo: Did you test with the latest v3 of the board and what kind of problems did you have?
Thinking about buying this one since it has a very good price.

Yes, I have version 3 that I bought from Sparkfun. The included software wasn't very accurate. I instead used the Razor AHRS library which has better documentation and a very friendly calibration routine to improve the accuracy of the IMU.

[virror]

Thanx : )
But for 50$ its still a good price for the hardware only, so might buy it and reflash it.

[Krenzo]

Thanx : )
But for 50$ its still a good price for the hardware only, so might buy it and reflash it.

If you're comfortable doing that, then go ahead, but if cost is your primary concern, then there's the $10 STM32F3 Discovery board.

[virror]

Cool, but its way to big : p
A bit restricting when experimenting with VR stuff : )
Might build my own IMU board instead, will see.

[ynsta]

STM32F*Discovery board are very very affordable and they include a prog/debug probe as well that if bought alone cost more than twice the price. But they are big and the pin affectation is not very good.

It should be very nice to build a small board with a Cortex M0+ or M3, some good gyro, accelero and magneto meters and maybe a tilt sensor, an usb port, and a bluetooth module like csr ones.

I'm might be able to do the design of the board with kicad. For pcb fabrication eurocircuit does a good job in europe, but they only produce the PCB not the assembly and I don't have the space at home to have a reflow oven.

Does someone know an equivalent to eurocircuit in europe (france preferably) that also can do the assembly of surface mount components for an affordable price ?

[virror]

No cheap ones : /
http://oshpark.com/ is super cheap when it comes to boards, but they don't do mounting though.
You don't know of anyone or maybe there is some kind of French electronics forum you could use to look for private help?
Thinking about doing a board with the MPU-9150 3-in-one circuit with hardware sensor fusion, seems like a nice one.

[ynsta]

There is this small board that use the mpu-9150, I might buy one and connect it to a DVI cable (as there is I2C and a 5V inside) to make some tests.

http://www.drotek.fr/shop/en/153-mpu915 ... meter.html

[virror]

Looks like a nice little board there : )
Be sure to post any findings : D

[ynsta]

MPU-9150 board received from drotek

I'll test it with I2C wires of a DVI cable and i2c-dev kernel modules (it permits to access i2c of my nvidia board ports).

[virror]

Nice, looking forward to a report : p
I have designed my own boards with that circuit and will order them soon, hope they are good : )

[ynsta]

Wiring done, but I don't think that I will be able to test it before next week.

Maybe this evening, but after that I will be away for some days.

I'll keep you in touch of my tests and results.

First test I want to do is evaluating the maximum refresh rate attainable in RAW mode of each meter (I am not interested by DMP mode).

Your board is a MPU 9150 board only or did you but some other components onto it ? MCU ? usb ? bluetooth ?

I also want to design a board with a cortex m3 (STM32F103CBT6 for example), an usb port, a STM port for debug/prog, bluetooth, a good low cost IMU, barometer, a tilt sensor, a SPI PROM or Flash to keep some settings and some external IO pin breadboard compatible. A small FPGA or CPLD alongside would be very nice also.

[virror]

My board is with a small 8 pins Atmel 8-bit microcontroller.
The plan is to chain a bunch of them together using a I2C buss and make a IMU based arm VR thingy to make some Unity experiments with : )
The whole board is around 2 square cm.

[virror]

My board is with a small 8 pins Atmel 8-bit microcontroller.
The plan is to chain a bunch of them together using a I2C buss and make a IMU based arm VR thingy to make some Unity experiments with : )
The whole board is around 2 square cm.
Im curious about both the raw data and the hardware sensor fusion, if the built in fusion is good enough that would make it so simple, just send a bunch of rotations to the computer and apply them to some bones : )

Edit: Hmm, what happened here, why was it two posts posted and with different content as well? : p
Maybe someone can delete the first one, cant seem to do it myself.

[ynsta]

I'm now able to read all raw meters from the mpu9150 with the I2C of a nvidia GPU DVI port.

Benchmarks :
Accel + Gyro 1000 reads ~ 6sec
Accel + Gyro + Mag 1000 reads ~ 12sec.

It's not very fast, but still have to check what is the speed of my I2C (100kHz or 400kHz) as the MPU9150 can run at 400kHz.

Even if time is divided by 4, this device might be interresting for some applications but too slow for a good VR/AR system.

Also the magnetometer is not very handy to use, for each read :
write cntrl register to start one capture,
wait for data ready on a status register,
read 6 bytes

My DVI to i2c cable is
SDA : brown
SCL : black
GND : yellow
VCC : orange
HPD : white

1k pull-up on HPD.
EEPROM I2C AT24C08C @0x50 loaded with a valid EDID to force port activation by the gpu.

I attached my source code with one recording in csv format.

My source is derivated from this arduino code: http://www.i2cdevlib.com/devices/mpu9150#source

[Fredz]

Accel + Gyro + Mag 1000 reads ~ 12sec.
Even if time is divided by 4, this device might be interresting for some applications but too slow for a good VR/AR system.

I'd say it's not bad at all for a 30€ chip. The 83Hz frequency is in the same league than the 87Hz of the PS Move, and the form factor is a lot more interesting. Why do you think it would not be suitable for VR/AR use ?

[MSat]

Using the DVI I2C channel is pretty slick! Certainly makes it easier to prototype with.


I'd really like to experiment with Freescale's FXOS8700CQ (http://www.freescale.com/webapp/sps/sit ... FXOS8700CQ) accel+mag combo. It's a pretty neat part - 16-bit 3DOF magnetometer + 14-bit 3DOF accelerometer. Each sensor can run up to 800Hz individually, or 400Hz in "hybrid" mode. It supports both SPI and I2C interfaces, the latter is actually pretty cool because you can configure the board so that the part has 1 of 4 I2C addresses, meaning you can daisy-chain 4 of them on a single bus. They're also surprisingly cheap at around $1.50 a piece.

Unfortunately, I have yet to find a distributor that will sell them in any quantity under 1000. Freescale does have a development board (http://www.freescale.com/webapp/sps/sit ... 47FXOW8700) but it costs $100, and you can't buy the individual breakout. I'll probably buy one soon, but I'm definitely not spending $500 to buy the 5 I initially wanted

At 400Hz for each sensor on the chip, and the ability to daisy-chain them without requiring a uC for each, they might be well suited for a multi-node skeletal tracking system. Or with the mag sensor operating at an 800Hz sampling rate, possibly magnetic tracking

[virror]

Very smart to use the DVI port for i2c!
83Hz is a tad bit slow to use for realtime tracking, but if it turns out that the i2c was just running at 100kHz then 332Hz should be more than enough.
Even though Oculus is sampling at 1000Hz they are not updating the headtracking at 1000Hz but using extra points for prediction and stuff as far as i know, 100Hz should be enough to make it pretty responsive.
ynsta:Do you plan on doing any tests of the built in sensor fusion? Would be interesting to see what speed we can get from that as well : )

Daisy-chaining was my plan as well but with a small 8-pin microcontroller on each board, the extra cost there is just $1 each so no big deal.
If anyone else is interested in this as well maybe i can order some extra circuit boards.

[ynsta]

I'd say it's not bad at all for a 30€ chip. The 83Hz frequency is in the same league than the 87Hz of the PS Move, and the form factor is a lot more interesting. Why do you think it would not be suitable for VR/AR use ?

Time from tracking to image display should be lesser than 20 ms, and ideally more closer to 10 ms. If it already take 12ms to read the sensors, it will be difficult to stay under 20ms.

I will continue to experiment with this one for the moment, but I also will look to other chips, with spi interfaces maybe.

Next steps:
- fusion algorithms.
- temperature impact on drift.
- calibration.

[ynsta]

Very smart to use the DVI port for i2c!
83Hz is a tad bit slow to use for realtime tracking, but if it turns out that the i2c was just running at 100kHz then 332Hz should be more than enough.
...
ynsta:Do you plan on doing any tests of the built in sensor fusion? Would be interesting to see what speed we can get from that as well : )

I'll look at the SPI frequency this evening (GMT), I brought my setup at work where I have access to a decent oscilloscope.

For the fusion not for the moment as I read on a lot of forum that it is mostly usable for smartphone applications, but if I have time I'll compare results from different fusion algorithms and the one included inside.

[virror]

According to the datasheet the built in Digital Motion Processor can be run at 200Hz, giving a latency of 5ms, add to that another 5ms for the sensor data and the whole circuit would have a max latency of 10ms if i rad things correctly. The transfer might add another 1-2 milliseconds depending on the setup, and of course all this depends on if the built in sensor fusion is "good enough". Would be really neat if the built in stuff could be used and we could just send the values straight to the skeleton : p

Edit: Seems that in can actually run the i2c buss in 500kHz but i guess not many MCUs or drivers support that rate since its not standard.

[ynsta]

Good news I tried with my laptop (with a GeForce 420M and a HDMI to DVI adapter). I2C clock is between 90 and 100 KHz, and the 1000 samples are read in 5.2 sec

So with a 400kHz FM I2C this time might be reduced to 1.3 sec (if not limited by the mag capture wait).

[virror]

Great news! : D

[virror]

Anyone knows were this sensor can be bought for a decent price in Europe? InteliSense want $54 in shipping for a few gram of circuits
And in Europe the seem to take $26 instead of $17 as they cost from InteliSense, seems like a rip-off anyway