thanks for posting the 2018 rules. Good work - I’m particularly looking forward to hearing more about the Technical Challenges!
There is one thing I would like to ask you about, since I was expecting but couldn’t find an update on rule 4.3, and German Locals have been asking about it:
In 2017, colored landmarks were first introduced. However, during the world championships some of those landmarks proved to be quite a challenge for teams using camera vision, rather than helping them. Most importantly, red landmarks (RGB 255,0,0) were found to be heavily interfering with the orange ball. This may have been partially due to warm lighting conditions in the convention hall, but it posed an unforeseen and near-unsolvable challenge to all the teams of the RCJ Open league.
In the new 2018 set of rules, said paragraph 4.3 has been kept unchanged. Are you planning on updating this passage in a later draft of those rules? Are there going to be any other means of helping the kids in distinguishing the red landmarks from orange balls?
Since ball recognition is one of the most important things a soccer robot needs to be able to do, this question is rather pressing for our local competitions.
Looking forward to hearing from you!
Best wishes
Marco
Hi,
I have to admit that the landmarks where a real menace at first, but in the end it simply meant that we had to improve the algorithms that we used. If you haven’t already, convert the image into the HSV color-space. Another simple trick is to count the area of the color detected area (since the marker is larger than the ball, you can automatically reject objects with a larger area than X). Shape detection is also not a bad idea, but if you use opencv, I don’t recommend using any of the native algorithms (houghcircles etc.), but instead I recommend them to code their own algorithms (easily done acording to edges since circles have none ) Instead of looking at this as a barrier, think of it rather as a challenge
thanks for your kind reply, we appreciate your advice!
We’re having a look at this issue from a national organizing point of view. While the Open league is indeed the most advanced of all junior soccer leagues, ball detection via camera vision in itself is already a new challenge for teams entering this league. Making it more complicated by deliberately causing interference requires yet another level of intelligence for the robot.
I’m not so worried about teams competing internationally. However, the red landmarks can prove to be a frustrating factor for teams on the regional and national level. Here’s our concern in detail:
A team may prepare for competitions at school in cold light and manage to distinguish the red color from the orange one while - under warmer lighting conditions - fail to do so at the convention hall. Since ball detection is one of the main three tasks a soccer robot must be able to do, this scenario will result in a complete and unforeseen failure with a lot of money and effort wasted.
From our point of view, some solutions might be
to use some other color,
to use a darker shade of red, or to
excluding landmarks on regional / national level - which would be sad, since rules shouldn’t diverge.
With this in mind, I’m afraid we have to ask TC again for their plans with §4.3. Are you planning on updating this passage in a later draft of the rules? Are there going to be any other means of helping the kids in distinguishing the red landmarks from orange balls, or could you provide us with any other advice for national competitions?
Hello,
Soccer Open is not a simple category. Either the teams bear with it, or they can drop to a less complex category. Last year, my team leader came to me and told me that I should make a functioning ball-detecting program on a Unix system. I was 13. All I knew was some basic Arduino and really basic python. In the 3 months that followed, I had to teach myself some pretty hardcore Object-Oriented Python, Bash and then C++. That year the robot didn’t work. We where beat by a lot of pixy users (with the exception of one team). But in the end, we were much better off than them Yes, they won the second place, but they missed on 3/4 of the knowledge they could have gained.
It is a hard category, and it has to be made harder constantly. New Abstractions to robotics evolve constantly. In 2004 there was no “out of bounds”, pretty rightfully, because it was hard enough to have a robot that followed the ball, especially since it was coded in Assembly/C. Today we have Arduino, Raspberry pi etc., etc.
A the transition from light to open is hard, teams then take a pixy and want to ease off the hard work that really pays off in the end, and then beg organizers for the removal of landmarks (yet another reason for the removal of pixies). And honestly, I think that last year it was totally fair that some of the local competitions had them removed. But the longer the landmarks are going to be there, the more I think they should stay there.
I have no say whether you will or won’t remove the in your local competition, but I certainly think teams would miss out some very important struggles that should be there.
I am not saying that RCJ isn’t for “kids”, since I can consider myself as one, but I think that the rules shouldn’t be simplified to make a simpler transition, since the struggle is an important part of the learning process.
thanks for your reply. Zvono is actually proving my point by acknowledging the landmarks as being more like a challenge than an actual help. From what I understood, this was not what TC had intended for them to be…
In principle I can see the argument being made both ways (i.e. for and against the landmarks) and to that end I can announce that a change in the rules is coming which should address both of these.
) Instead of looking at this as a barrier, think of it rather as a challenge