Autonomous Driving Using Inertial Aided KGPS and Vision System Based on Precision Maps

摘要

Autonomous driving or navigation has been appealing to car drivers for a long time. It will relieve drivers from tense of driving and increase comfort, safety and driving fun, especially when cars are networked and information access and e-commerce are extending from office and home towards mobile platforms. With a lane-level map, real time kinematic GPS can generally maintain a moving vehicle within a specified lane. Inertial sensors (INS) extend this capability to where short GPS outages occur due to bridges, buildings and dense foliage. A vision system can also be used for lane keeping by automatically recognizing lane marks ahead. However both systems have some pitfalls. The integrated GPS/INS suffers from long outages such as caused by tunnels. The performance of the vision system is very much dependent on weather and road curvature conditions. This paper investigates the performance of each individual system in terms of enhancing autonomous lane keeping and seeks for optimal combination of GPS, vision system and map data for achieving this goal reliably. A precision map with a lane model is built using statistic approaches for the test areas near our facility at Palo Alto and at Stuttgart. A middle range Honeywell HG1700 IMU and Novatel OEM4 receiver are used and the DaimlerChrysler in-house developed vision systems as well as a vehicle control system built with a Mercedes E420 are incorporated.