c2aee86157b4a40b78132f1e71a9e6f1-Reviews.html

The paper presents a new online approach for solving POMDPs. The paper builds on some solid work, including POMCP, AEMS2, which are some of the state-of-the-art methods for this problem. The key new insight presented is to prune the forward search tree using regularization of the policy size. Another contributions is a performance bound on the value estimate; this is used in the algorithm to direct the search. The paper includes a number of empirical results, comparing with other recent POMDP methods (online and offline).