Hi Bernt,
Yes, I took it out one day last week, but when I got to the field I could not get it to establish a telemetry link. I was out there for about an hour and a half trying to get it to connect. I was able to open a shell to the RPI, and I could establish telemetry through the 915MHz radio link so decided the problem must be with the UART/Ethernet converter. It would get very hot to the touch, don't know if that's normal, but decided to just eliminate it and experiment with Mavproxy thorugh the GPIO pins. So back home to the bench for a week.
As you mentioned elsewhere here, Mavproxy runs up a lot of CPU usage. I reduced the stream rate both in Mavproxy (--streamrate 2 on the command line) and in the Ardupilot parameters I reduced the stream rates of all SR2 categories by half. In other words, all parameters beginning with SR2 (for Telem2 port on Pixhawk) divided by 2 unless they were already at 1. This brought the CPU utilization of mavproxy down to 20% to 30%.
So this morning I took it back out and tried again. Instead of the bigger motor I mentioned earlier, I decided to just try a bigger prop on the D3530. Thanks to that and about a 10 mph wind I got a successful auto takeoff. My plan was to fly an autonomous mission that flew a rectangle about 7 km in perimeter, while I just watched the telemetry and video on my laptop. The video feed from Motion was less than spectacular, very choppy, about 2 fps was all I was getting. A couple of times the video dropped out completely (viewing it in Mission Planner HUD, it just went back to the green ground/blue sky graphic). I would just do the Set MJPEG source again and it would pick it up again but of course that's going to have to be resolved so it doesn't do that, or at least so it will automatically reconnect.
I suspect that the video drops are because the RPI is still just overwhelmed with Mavproxy + motion even with the reduced stream rates. My current video setting is 320x240, 15fps. On the bench the CPU utilization of Motion was 30% - 50% when I moved the camera around to simulate flight conditions. But at least the telemetry link was solid throughout, it did complete the flight and auto-landed safely, I'm happy about that.
So my ideas for improving the video feed, maybe experiment more with Motion's settings, and look at gstreamer again. This might sound silly but I'm also thinking about 2 RPi's in the plane, one for video and one for telemetry.
