Abstract
Haralick and Elliott's full looking ahead algorithm [4] was presented in the same article as forward checking, but is not as commonly used. We give experimental results which indicate that on some types of constraint satisfaction problems, full looking ahead outperforms forward checking. We also present three new looking ahead algorithms, all variations on full looking ahead, which were designed with the goal of achieving performance equal to the better of forward checking and full looking ahead on a variety of constraint satisfaction problems. One of these new algorithms, called smart looking ahead, comes close to achieving our goal.