3D Robotics Comment by Chris Anderson on April 15, 2009 at 6:06pm

There are only three trees around Sparkfun. We landed in two of them! Had the contest gone any longer, I imagine we would have been three for three. (In defense of ArduPilot, it's worth noting that it never happened in autonomous mode. Jordi's defense, meanwhile, was that "there are no trees in LA so I'm not used to watching out for them" ;-)

Comment by William Premerlani on April 15, 2009 at 9:19pm

Chris,

Once again, congratulations.

Regarding your comment about "waypoint acquisition was the low hanging fruit for improvement"....I am not sure exactly what you mean and what the problem is, but there may be an issue with GPS latency.

Both Paul Bizard and I have measured the latency of the EM406. It is 12 seconds. If you were using the EM406 today, I am even more impressed with your achievement. I do not have information about any other GPS units, but I will soon. I have a and an EM408 on order, I will measure them as soon as I get them. The ET312 that I used on my previous boards had a latency of only 3 seconds. The fact that the LOCOSYS has a reporting rate of 5 Hz is a good sign, but it does not necessarily follow that it will have a short latency.

Louis Legrand is also doing a study of GPS dynamics on some of the same units I am looking at, plus some others. He will probably start a discussion on the subject when he has his data.

Bill Premerlani

Comment by bGatti on April 15, 2009 at 10:43pm

Bill, they hit 4 waypoints in 35 seconds, with a total deviation of 365 degrees. Yours is a excellent place to start looking for the "low hanging fruit", but if the gps were 12 seconds behind, that would place it about 2 waypoints in the rears of the actual flight.

The comment was mine. I'm not sure what Chris and Jordi came away with - and I don't have the datalogs - so I'm guessing based on being there. No fun buying a lotto after the numbers are picked. So I thought I'd guess at the issues, and stand by to hear what Jordi thinks.

Here's my premonition. The Ardu lacks a yaw gyro, and it was windy. The direction of error tended I thought to be very much downwind. As you can see from the waypoints, Jordi set WP2? to be well upwind of the required route, this I suggest was to avoid being blown out of bounds (ie over the building) on the northern crosswind to WP3? He also may have set WP2? so far outside the box in order to create an obtuse angle at WP3? so as to avoid overshoot in the direction of the road - this is a better reason. In any case, I think the nav had a hard time holding a course in crosswinds, causing it to miss some WP thresholds. The solution I guess, might be to either open the Waypoint threshold, or improve the yaw hold - which is difficult without another sensor (gyro).

On the flip side, it didn't appear to hunt on the pitch or roll axis, so the ir system was impressive. I'm not sure missing a few ridiculously tight waypoints in the wind is necessarily a defect in need of a solution, it just seemed an obvious place to look for improvements.

Ben

Comment by David Low on April 16, 2009 at 1:40am

seriously, the EM406 has a latency of 12 seconds?!

Comment by Noth666 on April 16, 2009 at 2:26am

@Bill: I am looking at some new GPS units based on MediaTek chipsets (do not have the model# right now) which should be 5V and 5Hz, 32channels and about a quarter of the size (or less) of the EM406..
So I was wondering how you measure this latency is there an easy method for this?
I have a bunch of the chips so I'd be curious to measure them since they will be used in UAV/UGV if suitable.

Comment by William Premerlani on April 16, 2009 at 2:34am

David,

Seriously, the EM406 has a latency of 12 seconds in course over ground. Here is the experiment that I did:
I "plotted" the sine and the cosine of the course over ground out on servos, and took the GPS for a ride. I would drive along long straight stretches, eventually cog would be exactly right. Then I would take a right turn somewhere onto another long straight road. The servos would be absolutely motionless for 12 seconds. Then they would gradually turn to the new direction in another 3 seconds.

Paul Bizard verified the 12 second latency with tests that he did.

For the SparkFun contest, Chris and Jordi were doing exactly what my test was: right angle turns.

Another test that I ran was to go to an empty parking lot and drive in circles: the servos indicated circles, but they were 12 seconds behind my actual direction.

I also tried using ECEF instead of longitude-latitude. Same result, 12 second latency.

I then tried simply computing course over ground by taking changes in pairs of position points. Exactly the same results. So this implies that the same 12 second latency is in first derivative of the longitude latitude information.

There was one test drive, the first time that I tried using computed course over ground, when things seemed to get better. I do not know if it was wishful thinking, or whether the latency depends on something, maybe the number of satellites in view. But I have not been able repeat that result, it is always a 12 second delay with the EM406. I have an EM408 and a LOCOSYS on order, I will test them as soon as I get them.

Comment by William Premerlani on April 16, 2009 at 2:39am

bGatti,

I agree with your comments. Given the tightness of the way points, and the winds, the performance of the ArduPilot is very impressive.

I would be very interested to know what GPS Chris and Jordi were using. If they were using an EM406, I am even more impressed.

Comment by William Premerlani on April 16, 2009 at 2:47am

@Noth666,
There is an easy way to test, with either Chris Anderson's ArduPilot, or with my UAV DevBoard.
Write a program to read the GPS, and plot out either the course over ground on a servo, or plot cosine and sine of course over ground on a pair of servos. Then take the GPS and board for a walk or a ride. After you reach steady state on a long straight stretch, make a right turn and measure how long it takes for the servos to respond.

Comment by David Low on April 16, 2009 at 6:11am

@william: was thinking to get a EM406, but a 12sec latency is really bad for an uav. once, i done a simple drive around test with FV-M8(EB-85A), the latency is 2 sec with 1Hz update rate.
maybe i shud try a right angle test with it :)

3D Robotics Comment by Chris Anderson on April 16, 2009 at 1:41pm

Bill, we were indeed using a standard EM406, but in binary mode at 56,000 baud (Which is what the stock ArduPilot 2.0 and 2.1 code uses). Would that explain the better performance?

Comment

• ‹ Previous
• 1
• 2
• 3
• Next ›
• Page