In recent years, ridesharing has been an important part of the urban transportation system with the advancement of Transportation Network Companies such as Uber and Lyft since they can leverage technology and connectivity to enhance system performance. However, the system is still hindered by the imbalance between supply and demand in spatial and temporal dimensions. One plausible solution is the introduction of autonomous electric taxi (AET) and the central controller can preemptively and optimally reposition taxis to high demand area. In addition, the controller needs to simultaneously consider the need of human driven taxi (HV) because the near future adoption of AET is likely to include both types. This paper develops a framework namely RL-AET for dispatching single and carpooling taxi trips, repositioning, and recharging a fleet of autonomous electric taxi with a multi-objective of minimizing customer and system-oriented costs. The framework makes use of optimization models and minimum weight perfect matching for dispatching single and carpooling trips for both AET and HV and a reinforcement learning model for repositioning AET between zones and recharging. An asynchronous solving algorithm is used to ensure the framework is computationally capable of real-world network applications. Our RL-AET framework’s capability is demonstrated on the Chicago Road network with 7,393 links and 2,514 nodes and compared with the Baseline Model strategy from traditional practices. The carpooling trip model helps reducing the number of taxis needed by 58% with an average capacity of 2.3 demands per trip. The repositioning model increases the number of matchings by three-fold compared to the Baseline. Ultimately, the proposed framework is able to reduce the total weekly wait time and number of cancelled trips by 75% and 72% respectively. This study can benefit ridesharing providers’ in decreasing their operational cost and the travelers’ interest in decreasing their waiting and pickup time.