In this paper we introduce a new decentralized navigation function for coordination of autonomous vehicles at intersections. The main contribution is a navigation function designed for vehicles with predefined paths that uses expected time to intersection for collision avoidance. In such way, deadlock situations are avoided. Different inertias of the vehicles are taken into account to enable on-board energy optimization for crossing. Heavier vehicles that need more energy and time for acceleration or braking are given an indirect priority at intersections. The proposed decentralized coordination scheme shows a significant improvement in energy consumption and in motion smoothness compared to traditional crossing with human drivers.