Bimanual manipulation needs robots to be sensitive on the grasp force which is hard to be accurately detected. This paper proposes RL framework for enhancing the grasp quality during the bimanual manipulation. This framework is based on finger configurations and its feedback. After that, the grasp quality is evaluated by the reward mechanism for the hands to determine strategies. There are 2 strategies, simultaneous and interleaved strategies, which will be determined in this framework to manipulate objects. In this paper, the contour and centroid of objects to the robot are unknown. Through the RL framework, robots can perceive hand-object relation and then optimize fingers configurations. The simulations and experiments showed that this framework can improve the success rates and finger motion accuracy.