Different types of autonomous vehicles vary in design manufacturing and technical performance, and there are many uncertainties and inaccuracies in the selection process. How to select a safe and efficient autonomous vehicles becomes a great challenge nowadays. This article addresses a selection problem of autonomous vehicles by building an uncertainty model in multi-attribute group decision making (MAGDM). The evaluation information is described by the proportional hesitant 2-tuple linguistic term set (PH2TLTS). Based on the axiomatic definition of entropy, a comprehensive entropy measure corresponding to PH2TLTS is constructed and combined with the entropy weighting method to determine the attribute weights. In addition, a PH2TLTS comparison method is proposed on the basis of symbolic function and distance measure, which can not only determine the priorities between PH2TLTSs, but also indicate the degree to which PH2TLTS is greater (or less) than another. Combining Elimination Et Choix Traduisant la Realité (ELECTRE) III, the selection of autonomous vehicles is obtained by the proposed proportional hesitant 2-tuple linguistic ELECTRE (PH2TL-ELECTRE) III model. Finally, the case study, sensitivity analyses, comparative analysis are undertaken for verifying reasonableness and feasibility of the presented model.