Arora presents a polynomial time approximation scheme for Euclidean TSP in R2 that finds a (1+epsilon)-approximation tour in time nO(1/epsilon). The scheme can be extended to higher dimensions and other Euclidean problems in which the objective function is a sum of edge lengths.
(From a FOCS '96 paper by Sanjeev Arora)