udacity / p3_implement_slam Goto Github PK

It looks that while this project is great on it's own, it lacks the emphasis on computer vision aspects of SLAM.

I would suggest extending this project to actually use feature detectors from real/simulated world environment to try to detect features from the world (rather than directly converting landmark information from the world and adding noise on top of it)

May be we even don't need to go for advanced feature detectors like SIFT etc. Just using something like simple blob detectors and converting pixel coordinates to robot coordinates using perspective transformation would also serve the purpose of describing the idea. (You can have inspiration from Advanced Lane Lines Project from Self Driving Car ND or Project 1 or Robotics ND)

Given that the quality and standard of other projects from same nano degree and also from other nano degrees, I feel this project has less thought upon and unpolished. Besides very similar content is already there on Udacity free course CS373 so feels like just copy pasted here without proper coherent link to the syllabus

Full Disclosure: I have not registered for the nano degree so I’m not sure if there are any hidden topics on feature detection and key point descriptor extractions using something in the league of SIFT etc and converting them to measurements of the robot (may be using RANSAC and pose estimation etc). But from the look of the materials in udacity site it seems not the case.

As the title says, the final two tests at the end re-use the poses and landmarks variable names. So, if you run the entire notebook and decide to re-run the "Visualize the constructed world" section, the landmarks and poses get overwritten with the test set values. Please change the test set variables to something else like poses1, poses2, landmarks1 and landmarks2.