Open EgalYue opened 6 years ago
For example, If we use Artifical Potential Field algorithm: We need to build our _Ftotal: F_total = Fatt(q)+ Freq(q)
The KEY is to build the attractive potential function Fatt(q) and repulsive potential function Freq(q) based on our condition number(which means connect the both functions Fatt(q), Freq(q) to our condition number),right?
I think Potential field works now! I build our F_total like this:
F_total = Fatt(q)+ Freq(q) + Freq_condNum(q)
Freq_condNum(q) is the condition number distribution which already shown above.
I tested for some cases: Marker position : (0,30)
1.Far away from the marker:
2. Close to the marker:
Next step I will try to present the Error with a figure which is easy to understand.
Following presents 2 paths, one is computed with Potential field method, another is computed with A* method.(Now only same steps)
Potential field: https://github.com/EgalYue/Mobile_Marker_Based_Navigation/blob/742d93d24f5c15ebe94b8246e061cac34f919a5b/python/pathFinding/comare_TwoPaths.py#L398
Follwing presents the distance error between fixed path and measured path
Further, I will make more test for different path steps and describe the errors in other forms.
Test case: Different steps Start(1.55, 2..05), Goal(1.55, 4.05)
Test case: Different steps Start(1.55m, 2..05m), Goal(1.55m, 4.05m) Marker position: (0, 3m)
Another form to describe the error
The graphs look cool, have you tried increasing the grid resolution?.
I am wondering why the error for the A-Start path is lower when it is in the center (that should be fronto-parallel), right now you are plotting the error of the Translation vector, I would also like to see the error in the rotation vector in degrees.
Now the resolution is 0.1m in a [3m x 6m] square area, i will increase the resolution e.g. 0.05m
Now i used the Euclidean distance as the measure to present the Error, which means,
Error = sqrt((x_fixed - x_measured)^2 + (y_fixed - y_measured)^2)
Because the center has a lower condition number, its closer to the marker.
And the figures above showed the Error of Euclidean distance, not Translation vector, So next step I will show the Translation vector error and rotation vector error
Test case: Anther form to present the Error(Euclidean distance between the fixed path and measured path at each step) resolution: 0.1m Different steps Start(1.55m, 2..05m), Goal(1.55m, 4.05m) Marker position: (0, 3m)
Low "mountain" represents the path using Potential field method, which has small error High "mountain" represents the path using A* method, which has big error
Further I will present the T error and R error and increase the grid resolution.
We should speak a little bit more about this. Do you have time on Friday afternoon?
Yes, what time?
2018-05-02 14:59 GMT+02:00 Raul Acuna notifications@github.com:
We should speak a little bit more about this. Do you have time on Friday afternoon?
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/EgalYue/Mobile_Marker_Based_Navigation/issues/9#issuecomment-385969123, or mute the thread https://github.com/notifications/unsubscribe-auth/ASlu3pj9CL8Kw_AGiZDnwKGGyqQVLUlpks5tua25gaJpZM4TqQ0j .
3pm
ok, see you at 3pm on Friday
Question??? Whichever algorithm we use, RRT or Potential field or whatever.
The idea for the Path finding form A to B is: A -> O : Descent along the largest gradient O -> B : Ascent along the largest gradient
Is this idea right ?