Speed up collision checking

3 views (last 30 days)
RoboTomo
RoboTomo on 11 Nov 2021
Edited: RoboTomo on 18 Nov 2021
I am using Matlab's "checkCollision" function with four additional collision objects and a UR5 robot. Function is working great and returns logical 1 or 0, which is later used in the program (I am checking overall reachability of the workpiece).
The only problem I am facing is computation time. I already limited points on the workpiece but it still takes too long, 80% of the time is used only for collision checking.
Is there a way to speed things up? If not, is there any other available collision checking library (maybe with use of voxels or spheres) which works with Matlab in similar fashion?

Accepted Answer

Karsh Tharyani
Karsh Tharyani on 11 Nov 2021
Edited: Karsh Tharyani on 17 Nov 2021
Hi RoboTomo,
Thanks for your question.
  • There were some improvements made in performance for checkCollision in R2021b. Try and see if it meets your expectation.
  • You can also try turning "off" self-collision checking if you are confident that the configurations you pass to the collision checking routine don't result in the robot colliding with itself. See ("IgnoreSelfCollision") here: https://www.mathworks.com/help/robotics/ref/rigidbodytree.checkcollision.html#namevaluepairarguments
  • Alternatively you can try generating a MEX for an entry point function (See rbtCheckCollision) that accepts a configuration and returns whether the robot is in collision.
function isColliding=rbtCheckCollision(config)
% Use persistence if the robot and the environment are not changing.
persistent rbt env
if(isempty(rbt))
rbt=importrobot("universalUR5.urdf","DataFormat","row");
end
if(isempty(env))
c1=collisionBox(0.1,0.1,0.1);
c1.Pose=trvec2tform([0.2,0.2,0.4]);
c2=collisionSphere(0.3);
c2.Pose=trvec2tform([0.4,-0.2,0.4]);
env={c1,c2};
end
isColliding=rbt.checkCollision(config,env);
end
>> codegen rbtCheckCollision -args {zeros(1,6)}
I would be glad if you could also reach out to Technical Support and convey any performance requirements that you have for your use case, and we would be happy to enhance checkCollision in a future release of MATLAB.
Best,
Karsh
  5 Comments
Karsh Tharyani
Karsh Tharyani on 17 Nov 2021
Hi RoboTomo,
I apologize for the confusion. Yes "the input 'universalUR5e' did not match any of the valid values" is an expected error. I updated my answer, accordingly.
If robot imported via "importrobot" provides you with collision data useful for self-collision checks, please continue to use it as it is also supported for code generation, and you will be able to generate a MEXed version of rbtCheckCollision. You can verify if the imported robot has collision data useful for self-collisions, by using show
rbt=importrobot("universalUR5.urdf");
show(rbt,"Visuals","off","Collisions","on");
Question: Are you able to see collision data for the imported robot?
There seems to be an issue with "loadrobot" for the "universalUR5" wherein the populated collision geometries aren't useful for self collision checking. I have reported this issue to the development team at MathWorks. Should the issue be addressed in a future/update release of MATLAB, you shall be notified.
Please reach out to Technical Support if you are still facing issues, or have clarifying questions.
Best,
Karsh
RoboTomo
RoboTomo on 18 Nov 2021
Edited: RoboTomo on 18 Nov 2021
Ok thank you for clarification, I was already thinking that there was new UR5e model somewhere. Yes, visuals are working great, I was using this at the beginning to verify objects and collision detection.
I modified my program a little bit and separated collision detection. For calculating one configuration it now needs approx. 0.01 seconds which I think is good enough.

Sign in to comment.

More Answers (0)

Products


Release

R2021b

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!