problems with custom values of npackets

[bricerebsamen]

Hi

I just started replacing my custom velodyne driver with this one so that I have less code to maintain ;)

I am trying to publish one packet at a time so that my algorithms can be more incremental, as opposed to having to wait for a whole spin. So I am setting npackets=1. (I am using a 64E_S2 model)

That leads to 2 issues:

1. the diagnostic reports that the frequency is too high: 3472 vs 10

2. when visualized in rviz, with a decay time of 0.1, which should show a full spin, I see only 3 small fractions of the environment that are slowly rotating counterclockwise. I can't explain it, the publishing rate is fine (3472Hz on both velodyne_packets and velodyne_points). I wonder if you can reproduce this and/or explain it. With npackets=15 it seems to vanish or at least less noticeable.

[jack-oquin]

Hi Brice:

Problem (1) looks like a bug. I'll take a look and see if I can find it.

Problem (2) may be a matter of bandwidth, 3472 messages per second seems like a lot. Can your custom driver really handle that many without overruns? Fifteen packets per message seems reasonable, it is still pretty incremental. Have you tried running your algorithms as nodelets in the same nodelet manager process? That should reduce message overhead significantly.

I don't have a 64E S2. If you send me a PCAP file, I will try to reproduce problem (2).

[jack-oquin]

For problem (1): in driver.cc line 61 there is an obvious bug. The frequency computed is in revolutions per second. That is then set as the diag_max_freq_ and diag_min_freq_. That needs to be adjusted by the actual value of npackets.

[bricerebsamen]

yes my custom driver can handle sending single packets.

However it seems to be more rviz related than velodyne: the topic publish rate is reasonable, so it does not seem like messages are dropped.

My custom driver only runs in fuerte, so I can' test it with rviz in hydro. Rviz has changed a lot in between.

If you don't have a 64ES2, maybe you can try to reproduce with another velodyne model? I will also send you a pcap.

[bricerebsamen]

here is the pcap file: http://cs.stanford.edu/people/rebsamen/vel.pcap

[bricerebsamen]

btw, the vdump exec is not installed in hydro. I had to copy and paste the script from github to run it.

[bricerebsamen]

I tested under fuerte using the fuerte version of your driver and npackets=1. It's all fine. Seems like the issue is with rviz in hydro. Let me know if you find something, meanwhile I posted an issue on rviz: ros-visualization/rviz#689

[jack-oquin]

OK, I'll use this issue to fix problem (1), and ignore problem (2) for now. We'll see what the rviz developers decide.

[jack-oquin]

Since problem (1) is clear and straightforward, I opened issue #19 to specifically address it.