2
$\begingroup$

Gradient in Maximum Entropy IRL requires to find the probability of expert trajectories given the reward function weights. This is done in the paper by calculating state visitation probabilities but I do not understand why we can’t just calculate the probability of a trajectory by summing up all the rewards that are collected following that trajectory? The paper defines the probability of a trajectory as exp(R(traj.)/Z. I do not understand why we have to solve MDP for calculating that.

$\endgroup$

Your Answer

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

Browse other questions tagged or ask your own question.