Trajectory segmentation into curved intervalls
I plan a trajectory for a 6-dof industrial robot arm, then visualizing it in rviz. Other than testing different OMPL planners, I have left every setting at default.
Why is the trajectory segmented into smooth curved intervalls? Within obstacle region these intervalls get smaller, but why do they exist at the first place? Connection of these intervalls is done with a geometric "jump", i.e. the next segment starts in a completely different direction. I have read about configuration parameters like the longest_valid_segment. Please correct me if I am wrong, but this setting only affects collision checking in configuration space, so I dont see that it is interrelated with the described phenomenon.
EDIT:
I uploaded an example picture on github.
Can you clarify a bit what exactly do you mean by "curved intervals"? OMPL, using RRT, will plan in joint space. I assume you are familiar with the fact that linear interpolation in joint space will result in curves in a Cartesian space.
For clarification, I added a picture of the trajectory in the original question. From the documentation I have found online on the OMPL website, I assumed that all planning algorithms in OMPL plan in configuration space ( == joint space (?) )?
There is also this thread related to this question, stating configuration space based planning for OMPL planners.
For non-redundant robots I believe you can say
joint space == c space. That doesn't change my initial comment though.c space != cartesian space(in this case).re: picture: please just attach that to your post. Try to avoid linking to off-site resources, as if/when they disappear, this question will have lost some of its value.
re: picture: Unfortunateley I was not able to: ">5 points required to upload files". I will add the picture as soon as I have control.
If RRTstar in this case operates in c-space, it is clear to me why the trajectory line in cartesian space is curved. However, I dont understand why the path is divided into several curved lines. It seems like there is some mechanism splitting the path up and providing RRTstar subproblems to solve, then afterwards stiching them together. I observed that in regions, where edges of geometric obstacles lie (a box for example), the path segments get more short.
Can you point me into a direction where this might come from? My first guess would be a planning adapter by the planning pipeline.