What algorithm to use for collision avoidance of a robot?

Hi all

For a robotic arm, to "Back up" when a human approaches it, what algorithm should I use? RRT seems to go around the object, but I need the backing up. How do I do that?

Answer 1

[Sam Chak]

Hi @Farzad Torabi

I'm unsure. Maybe you intend to tell the robot to "back down" or retract the arm?

Conventionally, RRT is an "offline" path-finding algorithm that falls under the motion planning category. It does not intelligently tell the robot to avoid the obstacle that suddenly pops up in front of them.

What the robot needs in such situation is a reactive algorithm.

Generally in the collision avoidance algorithm, you should do a collision velocity check and identify a reasonable warning distance threshold.

[farzad]

Thank you so much for your answer!! what I want to do is also depicted in this video. So is that reactive algorithm existing in Matlab?

https://www.youtube.com/watch?v=NCLlx0u3h8Q

[farzad]

Does this command work :

https://www.mathworks.com/help/robotics/ref/checkcollision.html

?

[Bruno Luong]

Or for real implementation in these videos https://www.fogale-robotics.com/

You need foremost a sensor. The avoidance algorithm is easy I do it without any toolbox.

[Sam Chak]

@Farzad Torabi, I have not used the functions from the Robotics System Toolbox.

From the description, the function can check whether each pose of the robotic arm is in collision with the 'static' obstacles in the environment.

The function is helpful for the robotic arm to navigate if the obstacles in the environment are known before the path planning task is carried out.

You don't have to think some complex algorithms to retract the arm. You just need the Arm to 'Sense' the presence of human (obstacle) in the immediate vicinity, then 'Think' where to retract the Arm (in the pre-trained trajectories) and verify that the selected one is collision-free, and finally 'Act' to move the joints to produce the collision-free poses.

@Bruno Luong, By the way, the FOGALE technology looks really 'smart and cool'. Thanks for showing that.