곡선좌표계 기반 자율주행자동차 실시간 경로계획에 관한 연구

Abstract

Human-friendly vehicle is one of key issues for the development of the automotive systems. The primary principles of the human-friendly vehicle technology are the improvement of driving safety and convenience. Intelligent vehicle technologies have played a leading role to achieve the ultimate goal in automobile safety and comfort. The eventual evolution of the intelligent vehicle depends on the autonomous vehicle technology. The development of autonomous vehicle requires the state-of-the-art technologies in perception, planning, and control. Among three basic functions of the autonomous driving, the planning is the key element that makes the autonomous vehicle decides behavior of vehicle by itself. In this dissertation, a real-time path planning algorithm that provides a suboptimal path for autonomous driving with avoidance of static obstacles is developed. The proposed planning algorithm computes an appropriate path based on a set of predefined waypoints. The predefined waypoints are used constructing the base frame of a curvilinear coordinate system in order to provide path candidates for autonomous vehicle path planning. Each path candidate is converted to a Cartesian coordinate system and evaluated if suitable with using obstacle data. To select the sub-optimal path, the priority of each path is determined by consideration of the path safety, path smoothness, and path consistency. For safer driving of autonomous vehicles, following 3 factors - curvatures, safety, and length of generated path are sequentially required to compute the target vehicle speed. The maximum target speed is limited by the predefined speed limit of road profile. The target vehicle speed is restricted by limitation of the lateral acceleration using maximum curvature of the path. The target speed is restricted again by the safety and the length of the path. In order to implement and validate the proposed algorithm, the autonomous vehicle A1 was developed. For the implementation of the autonomous driving system of A1, this dissertation introduces a network based autonomous driving system that consists of four layers: sensor-actuator layer, low-level component layer, In-Vehicle Network (IVN) layer, and high-level component layer. The low-level components directly interact with the sensor and actuators. And the high-level components integrate entire information from low-level components and then make a suitable decision. All the data for autonomous driving in A1 are shared through IVN layer. The autonomous driving system presented in this dissertation was successfully implemented and validated by the autonomous vehicle A1, which won the 2010 Autonomous Vehicle Competition in Korea.