Feedback on possible changes to camera rules

Hello everybody,

first off: This is not (yet) an official rule change, just something we were prompted to think about by posts here in the forum (specifically Draft rules for RoboCupJunior Soccer 2022 released - #10 by stiebel) . Should a proposal like this go forward we would try to run it as a technical challenge in 2022 and depending on how it goes maybe include it in the 2023 rules.

Omnivision cameras are no longer significantly expensive (fisheye lenses for smartphones are available at around 15US$ and will fit most camera modules). Therefore we are thinking about dropping the requirement for teams to manufacture their own mirrors to achieve 360° field of view.

At the same time to keep it interesting one idea is to allow 360° for only one of the robots and restrict the other robot to something like 140° horizontal FoV so that there is something to be gained from tactically looking the right direction and/or sending information from the all-seeing robot to the other one.

What do you think?

Allowing of commercially available optical components

  • Yes, please allow buying optical components (mirrors, lenses, camera modules with large FoV)
  • Don’t care either way
  • No, please continue to require teams make mirrors themselves
  • Something else (please post your suggestions if you choose this option)

0 voters

Restriction of one robot to a smaller field of view

  • Yes, go ahead, one 360° and one 140° FoV seem like a good idea
  • Yes but different: Please choose different fields of view for the robots (please post your suggestion below if you choose this option)
  • No, please let us have 360° on both robots
  • No, please restrict both robots’ FoV to less than 360°

0 voters

Any other feedback and/or ideas on this topic or other aspects of the rules are always welcome, we do read all it.

Best regards
David

1 Like

Hello!

In my opinion 360 FoV optical system is the only one data source that allows robots to navigate on the field and precisely position themselves. Thus if overall FoV (I am not talking about the camera here, because conical mirror gives 360 FoV with almost any camera) for one of the robots will be restricted, robot itself will never be able to detect all objects on field (both goals and ball) simultaneously. And since both goals are crucial for navigation, gameplay will badly suffer from such a restriction.
Communication between robots can not fully replace proper vision system. For example there are often a situation when one of the robots is out of the gameplay. In this case due to the lack of the data robot with restricted FoV automaticly will be in a losing position compared to the opponents.
And finaly talking about communication. I believe that even now good communication between robots and proper tactics using both of them gives huge advantage even with 360 FoVs.

So i think we should not make any restrictions on robots FoVs but give teams more ways to using omnivision systems (by dropping requirement to manufacture own mirrors)

Thanks.

Team DOT.

LIDAR is one alternative, sensor fusion (Kalman filters etc.) are another.
In RoboCup Major SPL Humanoid league the Aldebaran NAO robots have successfully solved the positioning task with a two cameras with a total horizontal FoV of 60.97°.

This is true; it is also why we are proposing this rule: It intruduces some of the challenge humans face in dealing tactically with limited FoV.

Using both goals for navigation is the easiest way to get positioning out of a camera but by no means the only way. The NAO SPL teams mentioned above do it reliably with 60° of FoV by using field lines, goal posts and the fact that robots do not teleport (i.e. Kalman filters mostly) and have been getting reliably positioning out of 800MHz and 256MB RAM for 10+ years.

Please specify what you mean by “suffer”. If robots cannot immediately go for the ball but need to tactically look around and position themselves first that would in my view be a change to but not a detirioration of gamelay. Generally speaking I (speaking only for myself as a referee here) would welcome changes that devalue high power high speed motors and increase the value of good perception and tactics.

It is not supposed to 100% replace 360° vision, but it can offer some things 360° vision cannot. For example if a robot occludes the FoV of Bot A but Bot B can see what’s happening behind that is something not even 360° Vision on Bot A would afford you opening up tactical advantages for effective communication.

While this is true good positioning can be achieved even with limited FoV, just with different methods (mostly Kalman filters) than 360° vision would use. But this advantage would be symmetrical, i.e. the same to all teams so no unfair (dis-)advantage to any team would occur, just the existing advantage of more reliable robots having better chances.

This is an interesting point, could you go into some more detail on this? We all haven’t built robots in a while and are not up to speed on the details of how communication is being used these days.

Dropping the self-made requirement is likely to happen even if FoV restrictions don’t get adopted.

Thank you for your feedback and I’m looking forward to hearing from you on the detail-questions
Best Regards
David

For completeness I’ll cross-post @stiebel 's response here as well. The WiFi/UWB-topic isn’t strictly about the camera but a couple other remarks are.

Hello again,
we had the Robocup Junior German Open a few weeks ago and we experienced a new situation: The golf ball is smaller than most omniwheels. The camera - mirror construction mounted on top of most robots could not see the ball, once it was close enough as the mirror could not be mounted high enough to get a better angle.

So if we think about dropping camera restrictions regarding mirrors and lenses, what about dropping any restrictions at all?
We then would see optical orientation with a 360° Camera and a precision-view for ball and goal-direction with a front camera.
And for those who feel the desire to calibrate even more cameras they could use two fish eye cams facing front and back instead of building a 360° mirror…

@stiebel thank you for the info and great visualization. I added it to the committee’s sheet with all the notes we took considering the changes to the camera rules for next season.

Thank you for your suggestion @stiebel!

There is a standard counterpoint that we normally get to suggestions like this (and let me tell you, there have been quite a few of them internally) is the question of “doesn’t this make RCJ Soccer a money game” to which we usually lack a good answer. Would you happen to have some ideas around that?

Thanks!

Hi Marek,
I absolutely agree that robocup shall not be a money game.
So let’s face it: the cheapest camera/object-detection is the pixy cam for 70$ (or Euro if you like).
with a 320x200 pixel for detection you are fine to see a golf ball when facing to it.
If you want to have omnivision you need a mirror. Even if you have a good one (for around 100$) you have to see the whole field with both goals and the center of the image is useless as it sees the own robot from the top. There are just to few pixels left to see the golf ball on the other side of the field or just near the robot. So you have to switch to a better camera and a perfect lense. We have a full hd cam with an intel myriad chip and a raspberry pi to achieve this task. It works, but it costs a couple of hundreds of Dollars to do so.
Having one pixy looking up to a self made mirror to see the two goals for orientation and a pixy looking forward to see the ball is far cheaper.
And sure, you can always get better results with better tools, but having more expensive robots does not lead to an instant win.
With every additional camera the hassle to calibrate it increases, so I am sure there will not be a “camera battle” in numbers, but there could be an downsizing of the cameras if there is no need for a 4k image to see all that you want.
When I asked my teams about this issue they they have spontaneously favored a two fisheye solution, one to the front, one to the back, both pixys. That solution would see goals, balls, robots and would cost less then one pixy and one expensive mirror.
Best regards,
Roland

Thanks for sharing your reasoning @stiebel!