I have been looking at various autopilots to add to my collection but I have one less common issue. I fly all winter and its not unusual for me to fly at -20C. I have been playing with the ardupilot code and looked into the devboard but now that I am getting into the stabilization aspect of the AP I am wondering if IMU's will even work without massive temperature drifts at cold temperatures. I have read thermopiles have difficulty getting a reading when snow is involved.

I assume an IMU would work so long as I let it temperature stabilize before flying (I let me heli's sit outside for 10min or so before I fly to keep the gyro drift down). My 2 IMU's I would be considering are either ardunio based IMU or the UAV devboard IMU but I am wondering if its worth purchasing either of these units if they will not function during colder temperatures. I am working to wrap my head around DMC and how they help eliminate drift but its all fairly new to me.

Opinions are greatly appreciated,


Views: 1269

Reply to This

Replies to This Discussion

It takes up to 15-30 minutes in order to stabilize chip temperature from 15C (cold car) and -20C (full winter).
Powering IMU up at -20C usually moves chip temp by 5, 10C.
So it drifts all directions, all time, also with altitude.
Only good gyro2accel mathematical filtering is the answer.

I do not think it is an insurmountable problem. Some of the gyros have onboard temperature measurement which can be used for temperature compensation. Using temp compensation and reference vector compensation (DCM, Kalman, etc.) should take care of the problem even at -20C
this is exactly what I am saying.
Thanks guys, I think for this purpose I might have to go with a "roll my own" solution. I guess temp compensated chips to near milspec range should yield me the results I need. I remember reading about old radio transmitters that used "ovens" on the circuit board to heat the chips to a known temperature to solve frequency shift during temperature changes.

Nearly half of my flying time will be in colder temperatures so I guess I'll have to look more into custom autopilots, as it looks right now I cant seem to find any <$300 AP's with temp compensated chips.
Temp drift is only one part of positional drift.
Another is integration drift.
So you need proper math, and spending too much time on temp stabilization is not going to yield satisfactory results.
I agree...I have done projects in the past where well designed code can make up for less then perfect hardware but I was under the assumption that temp drift would be the root of my problems. I dont know enough (yet) about the math involved with accelerometers and gyros but it looks like I am going to have to delve more into this if I want something robust and useful.
Hi Curt,

I think you could use either the arduino IMU, or the UAV devboard, at -20C, provided you let the temperature stabilize before powering up....

You could also use Chris Bosak's system, he has nailed down a great solution to the temperature issue. (Chris, is your system for sale yet?)

The DCM algorithm includes a PI feedback drift controller. Once the temperature stabilizes, the integral term in the drift controller exactly cancels the gyro drift, no matter what the temperature is. The reason is, in steady state, the average value at the input of an integrator must be zero. The inputs to the drift controllers are the cross products of reference vectors (accelerometers and GPS velocity) with IMU representations of the vectors. The cross products are driven to zero by the PI controllers, so the axes of the IMU are brought into exact alignment.

Thermal transient conditions are another matter. When the gyro offset is changing due to changing temperature, the PI controller is adapting to it. It does maintain lock, but there will be small attitude errors. Once the transients are over, the errors are driven to zero.

So, if you let the temperature stabilize before turning on the UAV devboard IMU, it should be ok. As a data point, I was developing the roll-pitch-yaw demo last winter, and was testing it by walking it around outside in minus 20C weather, and it worked fine.

Even if the temperature varies a little bit after power up, performance is pretty good. In my EasyStar, my board is mounted inside the the foam fuselage (which acts as a great thermal insulation) near a LiPo battery that gets hot and warms up the entire compartment, yet there is zero roll-pitch offset after a typical 20 minute flight is over.

Analysis of the DCM algorithm and PI controller shows that it is not a gyro offset as such that is a problem, rather it is the rate of change of gyro offset (measured in degrees per second per second) that creates attitude estimation offsets. The amount of attitude offset error, in degrees, is approximately equal to the square of the PI controller time constant in seconds, times the rate of change of gyro offset, in degrees per second per second. The time constant is around 5 seconds. So, for example, to generate a 5 degree error, you would need a gyro that was characterized by an accelerating error at the rate of 12 degrees/second in one minute.

The accelerometers furnish the reference vectors for the integrators, so ultimately they determine the accuracy. But once again, if you let the temperature stabilize before powering up, accuracy will be pretty good, since the accelerometer errors do not generate "drift".

But this is just my opinion, and I have a "bias".

Best regards,
'You could also use Chris Bosak's system, he has nailed down a great solution to the temperature issue. '
(Chris, is your system for sale yet?)'
In sale, running slow and cautiously,
in order to keep user support controllable at this time,
I am restricting it to rudder only EasyStar.
Basically I am doing tests towards full pro use at completely different price level.
I am doing some crazy things up there.
My recent tests are 13km mission full rain thick fog 30km/h wind (up to 80km./h with the wind, -10...10km/h against, I am surprised it worked so well navigating 80% of the time without reliable GPS course, relying on IMU).

Curt uses flying wing, and unfortunately I have no time to switch to those right now.
Hey ChrIs,

Are you takig pre orders? :) I would like to purchase a fully operational EasyStar UAV from you whenever possible.

after working tirelessly, My EasyStar fell from the sky tonight while in WP mode with ArduPilot running JPN's 2.4.5 firmware :(

Just before the flight, I ground tested the plane fine with GPS emulator. Excited, I uploaded a mission and headed to the field.

It was dark, cold and a bit windy. ( FunLIght glow stick finger nails make awsome lights for night flights :)

I launched the plane in the opposite direction of the first waypoint assuming it would make a U turn when switched to WP mode...WRONG.!

when I switched to WP mode the plane throttled up, leveled out, attempted to turn around and dropped from the sky.

It caught me off gaurd kinda, could have recovered had I switched to manual in time. Oh well, another EasyStar gone.
Hind sight being 20/20, I should have launched towards WP #1..

It seemed like the thermopiles lost track of where the sky and the ground were.

Do they work at night?

Anyone ever have this happen?

Another approach would be to make sure the components operate at room temperature using thermal control by means of passive components (insulation) and active ones (heaters). This is very light and has been working in Spacecrafts for more than 30 years. On Earth you also have the advantage that air can circulate :-)
Toby thermopiles work day or night, testing a UAS on a windy night with light sticks was probably the issue.

AttoIMU is operating beyond MilSpec for temp. So if you want something to handle the cold, or heat, call Dean.

Thank you for the lenghtly reply. I am pleased to hear this, I have gone the "reinvent the wheel" way in the past with projects and would love to use an off the shelf UAV board for my purposes (no need to rebuild what already works perfectly well). After seeing Bryan's posts regarding the devboard I may have to give it a try as well. I already have a pickit2 that I uses at home for general PIC projects and now that I know its compatiable with the devboard its a pretty cheap solution to my problem.

I might have to order a devboard and do some cold weather testing. My wing that will hold the autopilot has to seperated chambers inside. One has loads of air circulation and will remain very close to the outside temperature and the other bay is somewhat sealed. The 2nd bay holds the ESC and Lipo which should keep the compartment somewhat warmer. I am not fully read on the DCM topic and have much to learn but your -20C testing results are encouraging. The last thing I wanted was to drop $100-150 bucks on a bit of hardware only to find that I can fly 6 months out of the year.


Reply to Discussion


© 2019   Created by Chris Anderson.   Powered by

Badges  |  Report an Issue  |  Terms of Service