Sonar SLAM re-start

[ivangushkov]

Time estimate: 30 hours
Deadline: 09.April

Description of task:

We have a Sonar SLAM implementation which should be fairly operational. Given that Sonar SLAM is fairly well developed as a technology, we should reevaluate this solution for both TAC and RoboSub. Camera detection gives us troubles with objects that are far away. The Sonar might help there, presumably being more long ranged than camera algorithms. Exciting possibilities are also present for Sonar/Camera fusion in trying to detect and classify objects in the map. Another thing that can be looked into is solving for the lost dimension of sonar measurements (elevation along elevation arc of sonar scans) through batch optimization methods. This is a possibility that a few people have looked into, so it is out there somewhere.

Resources:
https://github.com/WaldumA/sonar_slam
https://books.google.no/books?id=atJqCQAAQBAJ&dq=ekf+sonar+slam&hl=bg
https://www.cs.cmu.edu/~kaess/pub/Westman18icra.pdf

• Look into our existing EKF Sonar SLAM implementation and evaluate how much work is left on that. Consider the possibilities of using other SLAM methods (WOLF graph SLAM, etc).
• Evaluate the possibilities of running camera-sonar fused classification of objects
• Evaluate the possibilities of solving for Sonar elevation of map through batch optimization methods, CV style.
• Get it running with our FFS and rest of stack for pool test.