MAV Localisation (Section D of IARC 2017 Plan)

[aditya2592]

What is the plan for using arena grid for localization? @ManashRaja You were saying rectangle fitting wont work, what was the reason for that?

[mpdmanash]

@aditya2592 I am working on grid localization only now keeping aside everything else. My approach in previous attempts might be wrong. Trying a better one right now.

[sam17]

@ManashRaja Please write a design doc/approach that you are taking now also explaining previous approaches. It would be nice to have it in an issue which can be subsequently shifted to wiki.

[sam17]

Has there been any progress on this or any approaches tried or thought of? If yes, please update it here always.

[mpdmanash]

Update: Grid based localization (mLOG) is working. Tested in simulator data. Will soon test on real quad. Writing fake GPS for APM as of now to not wait for pixhawk to come. So will test localization and position hold on IARC arena.

[mpdmanash]

Update: Fake GPS injection successfully tested on APM. :smile: Now dear little APM can also get position input from non-GPS sources and we can use APM's control algos to fly autonomously just as with GPS and waypoints..

[aditya2592]

Awesome. Do mention in Wiki how you did it

[mpdmanash]

Update: Now it even works for aggressive motion of MAV, which was not possible in the previous version. Finally, will test this one in real-world dataset to complete task D1 which had the deadline of 1st April. The delay was due to the development and testing of this modification.

[mpdmanash]

PS: the labels in the graph is exchanged. :sweat_smile:

[mpdmanash]

By aggressive motion, here is the roll and pitch vs frame number graph. The localization was running at 17 fps (cam input fps). so dtheta/dframe was high enough to give wrong results in the previous method.

[Gauraviitkgp]

I'm Currently doing Fisheye camera Calibration, Will complete the theory by today and by tomorrow hopefully I would be calibrating it with the actual fisheye camera

[sourishg]

@Gauraviitkgp you can try this out: https://github.com/sourishg/fisheye-stereo-calibration

[Gauraviitkgp]

Update: Done with the theory part of Fisheye Camera, After complete revision of ROS, I'll start how to run Ros node on quad's Raspberry Pi. @sourishg I saw the repository, confused on what is popt?

[sourishg]

@Gauraviitkgp it's for command line parsing - install libpopt-dev. Also, that repository contains code for stereo calibration, I guess you only need to calibrate a single camera so you should check this out too: http://docs.opencv.org/trunk/db/d58/group__calib3d__fisheye.html

[Gauraviitkgp]

@sourishg even after installing libpopt-dev, the cmake files are showing an error:- fatal error: popt_pp.h: No such file or directory

include "popt_pp.h"

                 ^

compilation terminated. make[2]: [CMakeFiles/DisplayImage.dir/calibration.cpp.o] Error 1 make[1]: [CMakeFiles/DisplayImage.dir/all] Error 2 make: *** [all] Error 2