2
$\begingroup$

I am trying to understand Bayes Filter. In its update step we have P(z_t|x_t) as observation model.

Slide 37: Bullet 5 of this lecture states "Likelihood of measurement is given by “probabilistically comparing” the actual with the expected measurement."

For example from the following data:

Observation: [0, 1, 0, 0, 0, 0, 1, 0, 0, 0] -> This what the robot sense

Actual World: [0, 0, 1, 1, 0, 1, 0, 0, 1, 0, 0, 0, 0, 1, 0]

And actual world we can say [0, 1] resembles to [white, black] tile and sensor recognize them with [White: 0.7, Black: 0.9] probability. I am not able to understand here how can we calculate this likelihood for each step ?

$\endgroup$
1
$\begingroup$

The likelihood is with respect to your predicted state not the world state (You will never know the world state)

For example, if you predict you will be at [0,0] but the measurement tells you [.5,.5] you can calculate the likelihood of that measurement given your predicted state. In that scenario, you can use a normal distribution or some more complex function to calculate the likelihood.

In your case of a white black tile vector you can possibly treat each entry as a binomial distribution then calculate the joint probability.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.