[Dprglist] RoboColumbus postmortem

David Anderson davida at smu.edu
Sun Nov 27 13:26:32 PST 2016


Howdy DPRG,

Great meeting and robot run up at Avondale last week.  Very relaxed, 
great location, lots of learning going on.   We should do this more 
often than just once a year.

I've spent some time trying to understand jBot's behavior.  As mentioned 
I've not taken the robot off the shelf for about 6 months and had to 
re-learn what all the flags and parameters do.   Such crappy 
documentation...

In particular, jBot arrived about 15 or 20 feet to the left of the first 
two cones, but found the origin cone dead on, each run. What gives?   
The magnetic declination in Denton at the airport, at last report, is 
3.6 degrees east.  I played around with that a bit at the meeting but to 
no avail.   All I could do was to make the navigation worse.

This week I've convinced myself that the root cause is that, somewhere 
along the line, the IMU, which includes a 3 axis magnetometer, has 
shifted off center of the robot by several degrees.  I believe it has 
moved physically, and needs to be remounted and re-calibrated.  I 
probably banged it on something taking it to Colorado last summer.

So the whole pattern was rotated around the origin.   Hence only the 
origin location was unaffected, which fits what we observed.   This 
explains why mucking around with the magnetic variation offset didn't 
help.  It could help on one leg, while making the other two worse.

On the plus side, the camera blob-recognition and the search algorithm 
worked like a charm, and were indeed the only things that saved the runs 
from the off-center compass based (poor) navigation.    Also jBot's six 
wheel design with light wheel loading proved easily equal to the 
terrain, even the rock pile where it got briefly stuck, and especially 
the wet ditch.   Batteries need to be replaced, but I knew that going in.

So back to basics.

Scott, I'm pretty sure your robot could have handled the ditch without 
the extra central waypoint.  And without that waypoint, it probably 
would have beaten jBot's time.  Just a thought.

Perhaps Scott, Doug, and John would like to add their own perspectives 
on their robots' performances last weekend.  I'm sure that would be 
helpful to all.

cheers,
David





More information about the DPRGlist mailing list