According to the book Probabilistic Robotics, "the Markov assumption postulates that past and future data are independent if one knows the current state $x_t$." This is a central assumption in many localization methods, where it's invoked to allow the current state of the robot to be accepted as a complete summary of the past. It is generally accepted in this context that the assumption is only an approximation, which doesn't really hold in practice — but the robustness of Bayes filters can be counted upon to accommodate for that.

My question is, does that robustness extend beyond localization relative to a map, and into odometry estimation itself? In other words, is it "alright enough" to assume that, given an odometry estimate $x_t$, the previous estimate $x_{t-1}$ and the next estimate $x_{t+1}$ are conditionally independent from each other? I would love to find any academic references arguing this point either way.



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.