How do modern open source autopilots compare to aerospace-grade IMUs?

I noticed that Digikey is now selling Honeywell's newest aerospace-grade IMUs, which cost $1,328 each (note that's just for the IMU; it's not a full autopilot). How do the specs of these aerospace IMUs compare to those we use here? Are they worth the extra money? 

In terms of overall approach, the Honeywell IMU seem very similar to modern autopilots such as Pixhawk 2.x and 3.3: they both have MEMS sensors with internal environmental isolation and temperature compensation.

As for the sensors themselves, I'm no expert on specs, so I'll just post the basics here, comparing the Honeywell sensor to the Pixhawk 3

On the face of it, the Invensense and ST sensors in the Pixhawk 3 appear at least as good, if not better. But I imagine that there are some other factors that may be more important, such as gyro drift and vibration filtering. The Honeywell specs in drift are shown here: 

Meanwhile the Invensense ICM-20602 sensor in the Pixhawk 3 gives its drift in different units: ±4mdps/√Hz. I really don't know how to compare those.

Finally, I'm sure that a lot of the performance depends on the software running on the Pixhawk boards, be it PX4 or APM, both of which use GPS to augment the raw IMU data to compensate for drift, along with a lot of other smart filtering. 

So for those IMU experts out there: how do you think these two approaches compare? Are aerospace-grade IMUs worth the extra money?

Views: 5550

Comment by Curt Olson on February 13, 2019 at 7:19am

I remember a georgia tech paper from years before the DIY drone movement started where someone built an autonomous self navigating aircraft around a 1hz gps.  It doesn't take much to get an airplane self navigating around the sky.  So yes, accuracy, resolution, repeatability ... all good things to pay attention to in any kind of robotics.  Everything always boils down to your use case.

I have been drawing the EKF solution conformally over an action-cam video (post process.)  Here is my favorite demo video of my system (MPU-9250 based with a lot of careful tuning): https://www.youtube.com/watch?v=Su3sL4JhcQc  You can see it's not perfect, but the system is designed to clearly show any little imperfections.

I think if anyone did that themselves with the ardo or px4 EKF code, they would be surprised that their systems actually are not as accurate as they imagine.  In other words, the roll, pitch, and yaw angles reported by the flight controller are wrong by 1 or 2 or 5 (or more) degrees [nothing to do with the resolution of these values.]  The end result is the horizon doesn't line up and track with the real horizon very well, the sun marker doesn't sit in the center of the sun blob, etc.  Oh and the actual error can vary widely at different phases of flight or depending on the type of flying you are doing.  Calibrating your accels and mags can help, but again, probably not as much as people imagine.  I haven't seen anyone do anything quite like my video above with apm or px4 ... someone please prove me wrong here!

I'm not here to suggest everyone or anyone needs to upgrade their IMU to a honeywell unit -- especially if all they want to do is hover their drone and pan around with their gopro to capture cool video.  I just thought playing hockey with your IMU as a hockey puck and having it work perfectly through the whole game was pretty amazing/cool (and very Minnesotan.)

Everything takes time but at some point I hope my university lab can get one of these new honeywell units and run it side by side with some other popular systems.

Comment by Thomas Butler on February 13, 2019 at 9:52am

Like Windows Vista the MPU9250 is obsolete as is the Pixhawk.

Try using a mainstream Pixhawk clone and/or controller with a sensor of today such as the ICM-20602; The best made at this time by Invensense (even though they have "newer" products).  In addition, one must consider rounding errors that can creep into the software calculations. Not too that the HW devices have engineered hardware low-pass filters in their housings (these can probably be tuned to the application) which the typical Pixhawk/PX4 does not. Except for the Pixhawk2.1 and a few other newer "racing" controllers, but the efficacy is questionable.  Pixhawk and clones are also limited by their sampling rate capability limited by the Pixhawk hardware and NOT the MPU6000 sensor. The ICM-20602 data output rate is 32khz (yes I know the what the spec sheet says).

Note too that the HW product is a dedicated IMU. DESIGNED in hardware AND software. The Pixhawk is using embedded sensors and must do its IMU computations in the flight control software. I DOUBT the ArduPilot software has had the level of engineering (R&D) put into the IMU portion of the Ardu software that the HW device has. It is adequate for what it needs to do.

in effect, the comparison of the Pixhawk to a dedicated IMU is like comparing apples to oranges and is a non-issue.

Finally, why waste $10k on the HW device? Of course you will need three ($30k) to verify they all work the same in case you get a tweaked one or a bad one. We know the HW has superior data output at its relatively low output data rate. Think of what else could be bought with that $$$; some Intel i9 PCs !!.


100KM
Comment by DavidJames on February 13, 2019 at 12:33pm

Tom,   One of the key parameters for an inertial navigation measurement unit for a navigation system like the Pixhawk, that is assisted with a GPS, is the angle random walk level of the gyros.    The HG1120s are not that good of an IMU compared to the MPU6000 or ICM-20602.

The MPU6000 gyros has an angle random walk of 0.21 deg/root(hr).= (0.005 *60/sqrt(2))

The ICM-20602 has an angle random walk of 0.17 deg/root(hr) = (0.004*60/sqrt(2))

MPU9250 has an angle random walk of 0.42 deg/root(hr) = (0.01*60/sqrt(2)

The ICM-20602 has the lowest gyro noise, followed by the MPU6000.    Both of them are pretty good IMUs compared to 0.6 deg/root(hr) of the best HG1120s.

I have measured the gyro noise levels on the MPU6000 and they really do have that low noise level on the gyros.    The ICM-20602 is even better.

I think that one of the reasons that the Pixhawk autopilot has been so successful has been the remarkably low noise level of the MPU6000s.      Autopilots based on the MPU9250 with the higher gyro noise would have twice the leveling error.

Comment by Tom Jakel on February 13, 2019 at 5:12pm
David,

ARW is not the end all be all of parameters that matter in an inertial navigation system. It certainly is important but there are many other IMU specs to consider.

As always which of the specifications are most important depends on what you are trying to do with the navigation system.

I am a lead System Engineer in the Honeywell group responsible for the HG1120 IMU, HG4930 IMU and n580 Navigation system. If anyone has an experiment or commercial need that they would like to demo a system for shoot me a message with the proposal.
Comment by Thomas Butler on February 13, 2019 at 7:34pm

"one of the reasons that the Pixhawk autopilot has been so successful has been the remarkably low noise level of the MPU6000s." ...  Well probably actually because the software/hardware is adequate to do the job and is CHEAP at < $200.

Could the Pixhawk (using it just as an IMU) be substituted for an HW IMU in a full size UAV is kinda the jist of the original blog post.

A flight controller/IMU made with the the ICM-20602 can surpass anything made by HW and the Pixhawk. PROVIDED the supporting sensors's hardware low-pass filter AND the software take advantage of the ICM20602's 32khz data rate which allows for a relatively high Nyquist frequency.  Unless of course the HW product IS using an ICM-20602 sensor in its IMU, but they probably wouldn't release that sort of proprietary information.

It is not fair to compare the MPUxxxx/ICMxxxxx gyro chip output (at 4khz) to the integrated/massaged/hard low-pass filtered output of the HW device (at 200hz or so)...apples n oranges!

HW got rid of its 5883 mag sensor line (sold design to the Far East [and cloned by other Far East mfgr's with probably stolen fab files]), so one would conclude they are NOT actually making MEMS gyros (doing low-level fabrication) used in their IMUs .

BTW the monopoly China has on HW 5983/5883 compatible mags has now allowed them to hike the price substantially compared to five years ago! Good goin' HW!! Trump and Navarro need to talk to you about IP...

Happy flying!

Comment by Rafael Mendy on February 15, 2019 at 12:53pm

So

I'm confused after seeing the specification of Honeywell HG1120. The question is that what is the real advantage of this expensive imu with respect to many other much cheaper alternatives? nether ARW nor in-run bias stability are comparable to for example xsens mti1 imu. the price is one thenth. but performance is considerably better. So it means that ether these specs from xsens are false or there are other specs that i don't consider.

Can anyone help me understand?

Even in more commercial imus like mpu series it seems that the performance is comparable. however because they don't provide in-run bias stability factor, i'm cautious about comparison.

Comment by Thomas Butler on February 15, 2019 at 1:51pm

@Rafael, The difference is one is an AHRS IMU and the other are discrete sensors.. Regarding Pixhawk vs HGxxxx the difference is one is a purpose built AHRS IMU device; the Honeywell IMU, The other (Pixhawk,et al) is a controller that does its own IMU AHRS functionality/computations using discrete sensors(MPU6000, etc.). The HW product no doubt HAS discrete senors inside it. Who's sensors or what sensors is an unknown. Note too the relatively low bandwidth of the HW devices which provides for significant time for integrating (averaging) of the sensor data.

An interesting comparison would be to compare the specs of the actual MEMS sensors used in the HW device; i.e., raw data output.

Companies making devices costing upwards of $10k no doubt want to protect their market niche and not admit that the five dollar MEMS sensors are indeed getting better and can be just as accurate with the addition of good firmware. That does NOT mean the firmware in the Pixhawk is as good as the HW product; doesn't need to be to fly a quad-copter.

In part, it is a matter of perception, where >$ implies a "better" product. The HW device is an AHRS. It could be used as input to the Pixhawk via the CAN bus and the ArduPilot firmware could then read the HW device rather than do its own on-board sensor computations. The HW device probably has had considerable professional PAID engineering of the software which is something to consider; probably at least $500k.

Additionally, the $10k is paying for a software and hardware development process where every change is documented and justified and tested for efficacy. There is a reason GM goes to Cummins for diesel engines; it's their thing!

As far as "performance is considerably better", one cannot go just by the sensor's technical specs. How well the sensor/IMU filters noise to allow usable sensor data at the required minimum bandwidth and what is the true bandwidth of the AHRS system is critical.  The latest flight controllers can take advantage of the ICM20602's 32khz bandwidth to filter out a significant amount of noise from vibration.  Nobody is going to take a $10k HW IMU and use it as input to a Pixhawk on a multi-rotor that costs $1500. Note that the MPU6000 chip in the Pixhawk has a 32khz output capability, but cannot be utilized due to the Pixhawk's CPU to MPU6000 interface design.

Getting a bit political here now, but that $10k cost is also influenced by the volume of sales which is probably quite low which they have to amortize all the development cost over units sold. Five engineers at $120k each (that's total cost to employer) for a year plus facilities/equipment at about$100k plus product fabrication is about a $750k investment. Not chump change! The $$$ has to be recovered with a profit. The Pixhawk has had virtually NO software or hardware development expense (done for free by ZTH in Zurich Switzerland and other volunteers ). Only fabrication costs. Which amounts to about $50/unit from China and sold for $250 by 3DR (note that 3DR had virtually NO development costs and STILL they fizzled out).

In addition, all of you folks WHINING about the "high" cost of a $200 Pixhawk, remember YOU did NOT pay for the software OR the hardware engineering. All you paid for was the $100 cost of fabrication (by children no doubt making 25 cents a day) in and shipping from China. As a software/hardware engineer, it is a bit revolting when a person whines about the cost of a $10 BEC! It SHOULD cost $50. Trump's China tariffs should be 1000% to be fair to American manufacturers.


3D Robotics
Comment by Chris Anderson on February 15, 2019 at 2:08pm

Thomas, you're mistaken on the development costs. Both the hardware and the software teams that brought Pixhawk and its software to market were 3DR employees or contractors at the time. Total development costs were in the millions of dollars. Most of the key Ardupilot team members were paid by 3DR for more than three years (2012-2015). This is true for most open source projects, such as Linux -- the vast majority of developers are paid by some company to contribute.  The reason Chinese companies can sell clones for less is that they didn't bear any of those development costs, since we open sourced both the hardware and software.

Oh, and 3DR is now bigger than it was in the Pixhawk 1 days. We just don't do hardware anymore. It is, as you note, a tough market to make a profit in.

Comment by Rafael Mendy on February 15, 2019 at 9:34pm

@Thomas, I think HG1120 is not an AHRS but an IMU unit. it means that it's output is not Attitude data but raw or processed data of angular rates and accelerations ( most processed outputs are delta angular rates and accelerations )

As i said before, it's better to compare an imu product with similar outputs to HG1120. for example xsens mti1. What is the advantage of HG1120 to mti1? Xsens is a commercial ( not open source ) company as well. according to their datasheet, the performance of their device is better than HG1120. The price is almost one tenth. what is the difference in your opinion? 

Comment by Tom Jakel on February 15, 2019 at 9:38pm
The specifications of so many IMUs that you are quoting are at room temperature only, the Honeywell devices specs are over the full temperature range from -40 to 70 C. There are so many parameters that are calibrated for in the Honeywell product.

It is so difficult to make the comparison that most professionals who are users simply get a trial from the IMU company and compare by using in their application to see what works or doesn't and make the decision based on what they observe

Comment

You need to be a member of DIY Drones to add comments!

Join DIY Drones

Groups

Season Two of the Trust Time Trial (T3) Contest 
A list of all T3 contests is here. The current round, the Vertical Horizontal one, is here

© 2019   Created by Chris Anderson.   Powered by

Badges  |  Report an Issue  |  Terms of Service