this could become the most important channel in the field of education on youtube. its btw a great respect to the reasearchers, which are working for cyrill, that they are able to show their work.
The simplicity and clarity with which Prof. Stachniss explains the concepts sows your brain with new ideas. I'd watch these lectures a few times if I was thinking of research or project ideas.
Impressive how the lecturer makes sure to take the listeners along in every step of his explanation! Great example of a researcher dedicated to passing along his knowledge imho.
Great video! really enjoyed it thank you very much! i had been using a particle filter for my kinect localisation technique but had not known the name, this really helps clear things up! looking forward to your next video!
Still not sure on how the weights are being computed , like how are the probability distributions of the model state propagation or the observation model ? A bit not able to visualize the distribution , is it tabular form or ? thanks in advance !
Very impressive and well-explained professor, Thank you so much. So do you have suggestions for a preferable approach for highway lane matching localization between ICP & Particle filtering? Is there any specific advantages or disadvantageous over each method?
Thanks for the super video! From 30:35 you started explaining MCL. What is the distribution of p( z | x, m) ? Is it Gaussian, Uniform? Or does it depend on the problem? What distribution do people use as a likelihood ( p( z | x, u) ) in most cases? Thanks for your attention!
In the context of MCL, you don't really need to concern yourself with what kind of distribution p(z | x, m) assumes because you're not sampling from it; rather, you explicitly evaluate its value for every particle and use it as weight. For further info on observation model distributions however, you can refer to Ch.6 of "Probabilistic Robotics".