An autonomous car can operate itself without any help from drivers through intelligent control systems. Autonomous car needs a careful design of the autonomous driving system to proceed key functions such as localization, perception, path planning and vehicle control. Autonomous driving system is similar to human brain. As we make decisions based on information obtained through our sensory organs, the autonomous car is operated based on data acquired through various sensors such as cameras, lidars, radars and GPS. The system architecture of the computer controlled self-driving system is the most important part of the autonomous vehicle

Distributed real-time control system architecture for more efficient and reliable computing capability

Autonomous driving system consists of four core technologies: localization, perception, path planning, and vehicle control. Firstly, “localization” provides precise vehicle position information. Secondly, “perception” recognizes environment and assesses situation on the road. Thirdly, “path planning” provides safe path to destination based on information from localization, perception, and other situation. Finally, “vehicle control” manages the car safely to follow the generated path and to avoid any accidents. To drive the autonomous car safely and reliably, all core technologies must operate appropriately, even if the car is in dangerous and unexpected situation. Currently, most industrial and academic research on designing the autonomous driving system is based on the centralized architecture, which does all necessary functions with one or a few computer control systems.

Control system mounted on an Autonomous car. Efficiency maximized by occupying less space.

However, the centralized system consists of a complex hardware system to accommodate all necessary sensorsand numerous actuators. It requires huge hardware systems and additional electric power and high performance computing unit to provide a high computational capability.Furthermore, it is hard to guarantee reliable fault tolerance, because handling malfunctions and building backup systems are not easy for the centralized systemarchitecture. To overcome these drawbacks, Professor Sunwoo developed and implemented a distributed control system into the autonomous vehicle platform. This meanscontrol elements are allocated to several sub-processing units instead of one large control system. In other words, several processing units independently work for subfunctionsof localization, perception, path planning, and vehicle control of autonomous driving. Especially, the processing units are able to transmit and receive data over 100 times per second through the high-speed in-vehicle network. The system enabled the downsizing of processing units and the advancement of the control system.

Self-developing artificial intelligence system

The distributed control system can acquire and process relevant information much faster than the centralized architecture control system. Although errors occur, it can also easily diagnose

and deal with them. Hardware production costs as well as hardware maintenance costs can be significantly reduced compared to centralized architecture control method; resulting in improved efficiency of the Autonomous car. In addition, the computing units are suitable for easy plug-and-play. Therefore,any additional functions can be easily implemented. The development of the distributed control system advanced the level of safe-driving of the current Autonomous car even inhigh-speeds of 170 km/hour. The distributed control system can acquire and process relevant information much faster than the centralized architecture control system. Although errors occur, it can also easily diagnose and deal with them. Hardware production costs as well as hardware maintenance costs canbe significantly reduced compared to centralized architecture control method; resulting in improved efficiency of the Autonomous car. In addition, the computing units are suitablefor easy plug-and-play. Therefore, any additional functions can be easily implemented. The development of the distributed control system advanced the level of safe-driving of the currentAutonomous car even in high-speeds of 170 km/hour. The computing units comprising the distributed control system are developed according to each purpose. The core technologies of the distributed system architecture have two main parts: taskallocation and timing analysis. The task allocation distributes data processing load properly to individual computing units to prevent overloads, whereas the timing analysis is a way to confirm that all functions meet the required execution time in running. Through the system, it creates feedback of the driving process on its own and self-develops algorithms to enableoptimum performance of each computing unit.

Several computing units that are responsible for different functions are integrated into one network with distributed control system

In the actual driving environment, many unexpected situations occur. Situations that cannot be handled by mechanical algorithms alone occur constantly on the road such as lowvisibility due to weather conditions, inoperability of GPS under overpasses and in tunnels, application of reversible lanes, and broken traffic lights. Autonomous cars based on the distributed system that is essential in quickly recognizing and responding to such unexpected situations will maximize the safety andconvenience of drivers in the future. At the same time, the system will greatly contribute to the popularization of human friendly vehicles.