The integration of autonomous electric vehicles (AEVs) into urban public transportation represents a promising solution for sustainable mobility; however, optimizing energy consumption remains a major challenge, as current management strategies are frequently evaluated only in static and idealized environments. This paper proposes and validates an advanced predictive eco-driving strategy based on a free-coasting algorithm, specifically designed for an autonomous shuttle bus (EasyMile EZ10 architecture). The novelty of this research consists in evaluating the algorithm through a complex co-simulation architecture (IPG CarMaker coupled with MATLAB/Simulink) on a realistic 4 km urban route, integrating severe dynamic constraints: payload variation (passengers) and the stochastic presence of 13 signalized intersections. Utilizing Vehicle-to-Infrastructure (V2I) communication, the algorithm dynamically calculates the distance to the next stopping point, proactively interrupting the motor torque demand and applying optimized regenerative braking in the terminal phase. The comparative analysis of the simulations, conducted at 30 km/h and 50 km/h, demonstrates that the predictive algorithm eliminates the inefficient “stop-and-go” behavior, significantly mitigates battery state-of-charge (SOC) discharge peaks, and considerably reduces overall energy consumption (kWh) compared to a standard reactive controller, while also improving passenger comfort.