Monday, 2 December 2013

And the winner is...

... officially Linda, the Lincoln robot. But effectively it is the entire STRANDS team, really. Five robots (unfortunately, Rosie of KTH was undergoing repair) competed in our RobotMarathon, all running more or less the same software (most of it is available open-source on github). The final score board looked like this:
Final Score Board

Linda came out first in terms of distance travelled. Karl was the robot with the longest uninterrupted time of operation. All robots achieved quite a significant run-time and distance. And this plot only represents the longest runs without any expert intervention required!
But these plots hide some of the complexity and truth. The robots, despite running the same software base, ran in different environments of varying complexity that hold different challenges for our robots. And also their respective longest runs were accomplished with slightly differing software releases and parameterisations adapted to their respective environments. Just to list a few observations:
  1. BobKarl and Werner were running in a semi-public environment, with students being able to interfere with the robot, while Lucie and Linda ran in labs, which were still populated and were not designed for the robot particularly. But Linda and Lucie didn't have to deal with nasty people messing around with the robot, deliberately trying to make it fail (as it apparently happened for instance with Karl who had his camera being forcefully moved).
  2. Our robots had some issues measuring the distance travelled correctly due to a bug in the robot's firmware (never use floats with single precision, lesson learned) . Linda had a solution implemented to overcome this problem, but there is strong evidence that the distance especially of Bob has been significantly underestimated. He might have even outperformed Linda would the Birmingham team have adopted the fix developed by the Lincoln people in due time.
  3. The environments differed significantly in size and structure. Long, straight routes clearly give a higher yield in distance compared to smaller environments that require a lot of turning of the robot. 
Linda being actually helped
 by two boys to recover
from a navigation failure.
Also, the whole software-base that made this achievements possible was developed collaboratively. KTH contributed 3D obstacle avoidance and robust motion planning, Birmingham developed the overall control including recovery behaviours, and Lincoln shared their vision-driven autonomous docking and recharging, to mention just a few contributions. And also the whole STRANDS team engaged in continuous testing and bug fixing over the past few weeks. This clearly shows: The purpose of this competition has never been was to have teams battling with one another (e.g. in contrast to RoboCup competitions), rather the marathon was milestone for our project and a great opportunity to disseminate our research to the general public. So, the real winners are
  1. Software quality in STRANDS (probably most importantly; all robots ran without memory leaks, software crashes, etc, for days! Only firefox crashed one robot by draining all its available memory.)
  2. The general public who learned about our robot(e.g. through BBC Radio Lincolnshire 27/11/13: iPlayer or mp3, and print media like Lincolnshire Echo 25/11/13 and the Daily Mail 30/11/13)
  3. The carpet manufacturers; running robot on the same route for a week causes serious wear and tear...
  4. and Linda, because after all the graph shows she won :-)
Our marathon system will now serve as a data collection tool, allowing all partners to collect long-term data sets, eventually contributing the overall aim of STRANDS: Robot that learn from and improve through experience.

No comments:

Post a Comment