Research On Key Technology Of Auto-Driving Using Fuzzy Logic Approach

The development of autonomous vehicles for urban driving has seen rapid progress in the past30years. Modern autonomous vehicles are capable of sensing their local environment, classi-fying the types of objects that they detect, reasoning about the evolution of the environment and planning complex motions that obey the relevant rules of the road. The ability to navigate autonomously in these complex situations is accomplished by combining a variety of technol-ogies from different disciplines that span computer science, electrical engineering, robotics and controls, and others.Autonomous Vehicle control system discussed in this thesis includes many overlapping and integrating systems and techniques. Given captured images from cameras and data re-garding the position, nature, and velocity of different objects and hazards surrounding the car from various sensors like GPS, Radars, Camera, and others. The Scope of this thesis is not discussing techniques to segment the objects and classify them into defined feature vectors that fully describe them.The first phase begins with processing Given N Feature vectors describing each object, each feature Vector contains data describing:i) Nature of the object (car, human, tree, obstacle, etc.),ii) The distance between the object and the test car (in the term of (x, y) coordinates),iii) The relative Direction of movement of the object (in term of (x, y) coordinates), andiv) The current Speed of the car under research. The time of possible collision, which is a key factor indicating Risk is calculated between each object and the car under re-search. By using different classifying techniques the risk and danger level of each object is ranked based on the possible collision time criteria. Techniques such as:K Nearest Neighbor (KNN), Neural Network (NN), Support Vector Machine (SVM), and K-means are used in this re-search and compared to verify the best approach suitable for the presented data set.The problem of how to handle the N feature vectors and integrate them all to help giving a correct and accurate decision is handled in the second phase by introducing the Data Fusion System that combines data from multiple feature vector to achieve improved accuracies and more specific inferences that could not be achieved by the use of only a single input.Combine the N acquired objects feature vectors in a F feature vector that integrate all need-ed data for a proper decision. Two different Fusion techniques are proposed in this research context:A) Data Fusion based on their relative position to the car under decision (Highest risk from (Right, Front, Left, and Behind)). B) Data Fusion based on the objects’nature (Highest risk of (Car, Human, Traffic Light, and Obstacle). The two proposed approaches are com-pared and the best approach is selected to be the input for the decision phase.The purpose of this Thesis is presented and covered in the last phase which will introduce Fuzzy Logic Decision Algorithm that will propose how to use Fuzzy Logic Control system to solve the problem of uncertainty regarding multiple choices and the proper decision. The idea of the proposed Algorithm is about behavior selection which is about depending selection of which distance affecting the collision time with testing car. The algorithm starts by selecting which behavior shall it take and this will be done "and" rule in the first line. It explains the car handling cases for every object which is front only, behind, left only or right only.The result will be in a form of a decision to the driver or the agent suggesting the proper direction of steering and proper velocity to avoid possible collision.