When multiple miniature vehicles with individual position and inter-vehicle distance measurement ability collaborate in a formation, navigation base can be established by data fusion in a decentralized and standalone scheme. A Composite Data Fusion (CDF) algorithm which combines Least Square Error and Kalman Filtering is proposed in this paper to build navigation base with optimized computing stress. In CDF, Enhanced LSE is incorporated as the preprocessing stage to build a coarse estimation and handle temporary or permanent group number failure. KF stage is then built to further alleviate noises in the pre-processed estimations In CDF, the dynamic model can be much simpler than KF, so the computation load is reduced while the result still has the advantage of high precision. Simulation results show that, when the fault rate of measurement in each vehicle goes 5 thousandth, the result is still acceptable. The computation time of the proposed method is less than three percent of that of KF, while its precision is almost the same to that of KF.