Traditionally Robotic grasping is performed by a robot manipulator with a rigid end-effector (a hard gripper). The usage of such kind of end effectors requires a high accuracy object detection to detect the best candidate contact points and precise control for the motion planning to perform a successful grasp. Additionally, force control is needed to prevent damage to the target object, especially if it is fragile. On the other hand, grippers made of soft materials shown to be compliant with different object shapes and can perform a safe and stable grasping due to the nature of soft material. In this thesis, we propose the implementation of an imitation learning algorithm to simplify the control scheme, by combining with a pneumatic soft gripper. As a result, the system proposed exploits such advantages of soft grippers so the limitations of a hard gripper based system can be solved. We compare the performance of the model to the demonstrations collected by changing the position, pose, and shape of two different target objects, and additionally, we tested the grasping successful rate.