Self-Driving-Cars-Specification


Autonomous driving is regarded as one largest application area for Artificial intelligence and it is this favored potential as well as this huge business values that researchers and scientists are willing to put much emphasis on it. It should be noticed that the cover image is from Google image. Basically, several specifications could be considered as representations to interpret its principles, which are related to control, state estimation and localization, motion planning. Control could be split into several parts as well, which method is utilized without complicated algorithms. For example, data is captured using sensors and then vehicles is controlled with one PID control method. When it comes to state estimation and localization, it could also be named perception and prediction. The operation of an autonomous vehicle requires actuation of 3 major controls that are acceleration, steering as well as stopping to keep the safe control. For perception, mapping, localization and detection are emphasized on the process after sensors receiving data stream. For prediction, either predicting vehicles’ or pedestrians’ trajectory would improve the whole driving efficiency, especially on the highway. For planning, it is one another popular area to avoid collision and keep safety. Motion planning could be divided into global and local methods, as well as behavior planning for possible multi paths with probabilities to reach one most optimized.

Apart from the theory i learned from tum, and some projects related to codes have been completed as well. For example, projects have been completed as follows: lane finding using simple computer vision(opencv) method, traffic sign is classified using TensorFlow, behavioral cloning is done using TensorFlow and Keras, advanced lane finding, especially lane curvature and vehicle displacement, identified, with the help of deep learning method, vehicles are detected and tracked. Beforehand, methods, most of them, are mentioned using python to program. Then, for tracking the bicycle’s position and velocity, Extended Kalman Filter(EKF) as well as Unscented Kalman Filter(UKF) are adopted. Similar with EKF and UKF, particle filter is used to estimate the future position of the vehicle. Then, PID and MPC controller are used to keep the vehicle stay in the middle of the lane. Another project related to program with c++ is that building a planner to create smooth and safe trajectories for the car to follow others when driving in the highway.