RoboMagellan 2012 Build Diary - Post competition analisys

Days 1 - 7

Days 8 - 16

Days 17 - 27

Days 28 - 34

Days 35 - 41

Day 42

Day 43

Day 44

The day of the competition - I did two runs, the closest I came to a cone was about 15 yards (from Cone 3).


I had a bug in my NMEA (GPS) decoding state machine.  The robot thought the cone was in China.  In the US, longitude is always negative (-68 = Maine, -123 = California ).  I forgot the negative multiplier in the code.  The robot was constantly navigating towards China.  Unfortunately, my GPS does not work at all in the house, so I did all of my GPS testing via simulation (hard coding constants in the state machine.  I had always correctly encoded the longitude constant with a negative sign, so I never realized the bug until I did more outdoor testing at the event.


The robot is now correctly navigating to GPS coordinates (here in the US).  I am currently investigating another bug that causes the robot to miss waypoints in its internal map.  This bug has been impossible to find without telemetry.   Today I added a Bluetooth modem to the robot.


What did work well on the robot was the ultrasonic sensors and software along with the closed loop speed control and motor control in general.  The robot regulated its speed on concrete and only had minor issues in tall grass.  Towards the end of the day, the battery was going dead so it got stuck a lot towards the end of the day.  I definitely need to add the Integral portion of the PI controller.  When the robot would go into tall grass sometimes it would stall for a second and then start running.  The Integral component will fix that.


The Ultrasonic system worked very well.  I never had any issues with bushes that were eating other robots.  This was sue to the two back sensors being aimed up.  The two back sensors see the tops of the bushes, much easier to see than the trunk or lower portions of the bushes.


I do need to consider adding breaks to the robot.  Even if the sensors detected an object, sometimes the robot would still hit it by simply coasting into it.  There was also a problem with object that were too close.  I think the MB1000's cannot see objects that are too close.


Thinks still left to do on the robot:


1) Fix the map bug

2) Finish the Raspberry PI vision system

3) Layout a board to cleanup the wiring mess

4) Integrate the power supplies

5) reconfigure the mechanical robot to be smaller



Views: 523

Comment by eric gregori on August 8, 2012 at 1:13pm

While my robot failed to touch a single cone (the best I did was coming within 15 yards of cone 3) these robots not only found the cone, but touched it.  In one case, the three time winner of this event touched two cones.  I have two bugs fixed and I am closing in on a third.  I am already looking forward to the next event.

Credit for the picture below goes to Salvador.   

Additional pictures can be found here:

Three time winner of the Chibot's SRS RoboMagellan event: Rick Brooks and NoName

What I find so incredible about No-Name is that it finds cones without computer vision.  It strictly uses

ultrasonic ranging and good programming.  You can learn more about No-Name and Rick's other robots here:

The CIRC robot is a model of quality engineering.  It's a very impressive custom mechanical assembly.

They are using a netbook as a controller.  It must be running Linux because if it was running Windows it would have never found the cone.

This is Steve's really cool Lego bot called Green Machine.  Who would have thought you could do RoboMagellan with Lego.  The robot has been very successful wining the first Chibot RoboMagellan and coming in third this year.  The Green Machine does a great job of finding cones, it just needs some faster motors from Lego.

This was the second coolest robot at the competition (after mine of-course).  This is the future of the competition, smartphone based robots.  Steve and his co-builder used an Android phone as the brains and sensors of the robot.  This robot crashed into a lot of things but it did touch a cone.  I thing there is a lot of potential in this concept.  There is still a lot of work to do, I look forward to seeing it next year.  This robot came in fourth.

Since I did not touch a cone, I do not get a place.  My robot is based on the very adaptable Minds-i ( robotics platform.  The robot covered rough terrain without any problems the closed loop speed control did a good job of keeping the robot at a reasonable pace on the concrete yet giving it all the power needed for the grassy areas.  The robot did a great job avoiding bush's but really seemed to enjoy crashing into people.  It also enjoyed crashing into posts.  Post analysis has found three software bugs so far, I am still chasing a fourth.  In my experience, the most important part of RoboMagellan is testing.  Test, Test, and retest.  A valuable lesson for next year.


Comment by Salvador Garcia on August 16, 2012 at 4:49pm

HI Eric, great post!

I got some video of one of your runs:

Sorry about missing the other run.

Also, credit for the images should go to Al Rossetti. My pics aren't up yet. My part was helping Al post the images and posting the direct link to the image gallery.

China, huh? I am glad you had that robot on a leash! Otherwise you would have been chasing it there!


You need to be a member of buildsmartrobots to add comments!


© 2017   Created by eric gregori.   Powered by

Badges  |  Report an Issue  |  Terms of Service