EuRoC Forum

[proprentnerl]

Hello!

The discription reads as if there is an infinite number of possible combinations of boxes and cylinders as objects in the scene.
Task 1 to 4 start with the same 3 objects red_cube, green_cylinder and blue_handle.
During the evalution will there be the same objects, will there be sligthly different or completely new combinations such as a box with a cylinder on top?
I also wonder about the varition in size and color of the primitives (e.g. red_cube).
Is there some kind of maximum to the side length?
Is it possible that there are 2 objects with the same color in the scene?

Thank you!

[fschmidt]

hello - sorry for the late answer!

the number of objects to grasp will stay the same in task 1-4. as our
focus is on shopfloor logistics where objects to grasp are typically
known we will only scale their shape between 25% and 120% of their
current dimensions from the current set of scenes (task*_v1*, from
euroc-c2s1-scenes_1.0.3_i386.deb).

there will be no other changes and no new (compound) shapes.

for all tasks (1-6) the objects color will always be selected from
this short set of 8bit red-green-blue hex color codes:
0000ff - blue
00ff00 - green
00ffff - cyan
ff0000 - red
ff00ff - magenta
ffff00 - yellow

in task 1-4 there will be no two objects having the same color (see http://euroc.iphpbb3.com/forum/viewtopic.php?nxu=20112706nx57588&f=4&p=409#p409 below)

beside the feasibility to grasp them there is no restriction on the
initial object pose.

for task 5 the number of parts is below 17 and their shape is not fixed. their
shape will always be a compound made out of equally sized cubes.
parts might have the same color but will then be clearly
distinguishable because of their shape and will not touch each other in the final configuration.
The size of the cubes will be between 60% and 120% of their
current dimensions from the current set of scenes (task5_v1, from
euroc-c2s1-scenes_1.0.3_i386.deb).
(please also read this post: http://euroc.iphpbb3.com/forum/viewtopic.php?nxu=20112706nx57588&f=4&p=375#p375)
the relative_puzzle_part_target_poses will have their rotational components always set to be 0:

Code: Select all

relative_puzzle_part_target_poses:
    part...: [ x, y, 0.5*boxsize, 0.0, 0.0, 0.0]


in task 6 there will only be the current red_cube objects but these
properties of the conveyor belt will change:
- move_direction_and_length
- drop_center_point
- drop_deviation

we will provide atleast one other set of scene files (task*_v2*) which
will demonstrate those variations.

to test the robustness of your solution against those variations you
can also create your own task-versions by copying one of the existing
task-description files from /opt/euroc_c2s1/scenes/task*.yml, modify
them and add the new filename to scenes.yml (within the simserv
VM). but please do not change the existing *.yml files because only
scores generated with unmodified official scene files can be listed in
the (optional) weekly leaderboad.

[edit 2014-10-30: task 1-4: there will be no two objects having the same colour, task 5: relative_puzzle_part_target_poses will always have 0 orientation, there will be no more that 16 parts, parts might have same colour but only with different shapes and "not-touching" relative_puzzle_part_target_poses]

[red]

for all tasks (1-6) the objects color will always be selected from
this short set of 8bit red-green-blue hex color codes:
...

objects might have the same color but will then be clearly
distinguishable because of their shape.
...

for task 5 the number of parts and their shape is not fixed. their
shape will always be a compound made out of equally sized cubes.

Concerning task 5:
It will lead to problems concerning object detection, if parts which are located next to each other in the (solved) puzzle have the same color.
Can we assume, that this will not be the case?

[red]

for task 5 the number of parts and their shape is not fixed. their
shape will always be a compound made out of equally sized cubes.

Can we assume at least some minimal size of the boxes in task 5?
If the minimal size is also 25% of the provided example, then the size of a single box (0.0125) is much smaller than the width of the gripper (0.03) and even smaller then 2 times the height of the gripper (2*0.0085).
A minimal size, especially equal to the the gripper width, would simplify the algorithms significantly.

[Peter Lehner]

Concerning task 5:
It will lead to problems concerning object detection, if parts which are located next to each other in the (solved) puzzle have the same color.
Can we assume, that this will not be the case?

Yes, you can assume that no touching parts will have the same color.

If the minimal size is also 25% of the provided example, then the size of a single box (0.0125) is much smaller than the width of the gripper (0.03) and even smaller then 2 times the height of the gripper (2*0.0085).
A minimal size, especially equal to the the gripper width, would simplify the algorithms significantly.

We have decided the minimum size of the boxes is 60% (~ 0.03 m) of the current v1 size for task 5: The boxes will not be smaller than the width of the gripper jaws.

[Tobias Fromm]

Hi guys,

just to make sure: Does this

objects might have the same color but will then be clearly
distinguishable because of their shape.

mean that in task 1-4 there may be more than one object of the same color, but different shape existing in one scene?

If yes, that would be contradictory to what the Technical Annex states on p. 19:

The objects field contains information about the objects important for the task. In task 1 - 4 these are a blue handle,
a red cube and a green cylinder.

Cheers,
Tobi

[fschmidt]

just to make sure: Does this

objects might have the same color but will then be clearly
distinguishable because of their shape.

mean that in task 1-4 there may be more than one object of the same color, but different shape existing in one scene?

in the Call for Challengers Annex III ‒ Scoring System for Stage I (EuRoC_Call4Challengers_Annex3.pdf page 11) it was stated

Task 1... each of different colour and shape

because of this we will add a new limitation that in task 1-4 there will be no two objects having the same colour.

If yes, that would be contradictory to what the Technical Annex states on p. 19:...

this section of the annex is an example of a possible yaml description as the section title suggests

2.4.3 Example Description

and not meant to limit the possible parameter values.

[Tobias Fromm]

Thanks for the clarification! That helps a lot.