A system and method for performing self-learning for adaptively achieving passenger comfort enhancement in an autonomous vehicle. The system comprises a plurality of sensor inputs. Each sensor input provides data representative of voice responses and image responses from a passenger in the vehicle. A controller is coupled to the plurality of sensor inputs. The controller generates and updates a reward function that includes a plurality of driving state transitions. The reward function is updated based on destination information and the data representative of the voice and image responses. The controller further generates a goal function that determines an optimized driving state transition updates the goal function based on the updated reward function and a previous goal function. The controller also generates a vehicle speed control signal, based on the updated goal function, to control the speed of the autonomous vehicle.