Study on a Fuzzy Q-Learning Approach Using the Driver Priori Knowledge for Intelligent Vehicles’ Autonomous Navigation and Control 2018-01-1084

The functional elements of decision making system are fuzzy, adaptive and self-learning for intelligent ground vehicles. As is well-known, operating environment of unmanned ground vehicles (UGVs) is complex, unknown and time-changing. And on the other hand, exact dynamic model of the vehicle is relatively difficult to gain. However, the changing of special dynamic parameters and the man-made driving laws of velocities and running direction are easily available. Therefore, this paper attempts to provide an approach based on fuzzy Q-learning algorithm for studying autonomous navigation and control system’s design, which aims to make unmanned vehicles adaptive and robust under complex and time-changing environment. The presented approach utilizes the drivers’ empirical knowledge for. Fuzzy inference system introduces the human beings’ successful experiences into the system, and Q-learning mainly pays more attention to the interaction between the robot and the environment and thus keeps on learning until achieving the goal. Through this method, autonomous navigation and control system can be designed accordingly. This paper used a type of the nonholonomic robotic system for the computational experiments so as to verify the algorithm, which only considers necessary candidate conclusions. The final simulation results show the validity of the designed algorithm. The presented algorithm is not dependent on the dynamic model, and is designed in terms of the special model parameters. Therefore, the mentioned approach has strong versatility and transplantable, which can be easily used for penetration maneuver strategies and autonomous maneuver of other type of intelligent vehicles such as unmanned aerial vehicles (UAVs) or autonomous underwater vehicles (AUVs).