On-line road boundary modeling with multiple sensory features, flexible road model, and particle filter

This paper describes a method of robustly modeling road boundaries on-line for autonomous navigation. Since sensory evidence for road boundaries might change from place to place, we cannot depend on a single cue but have to use multiple sensory features. It is also necessary to cope with various road shapes and road type changes. These requirements are naturally met in the proposed particle filter-based method, which makes use of multiple features with the corresponding likelihood functions and keeps multiple road hypotheses as particles. The proposed method has been successfully applied to various road scenes with cameras and a laser range finder. To show that the proposed method is applicable to other sensors, preliminary results of using stereo instead of the laser range finder are also described.

► We develop a road boundary tracking method for mobile robot navigation. ► Multiple sensory features from cameras and a laser range finder are integrated for robust estimation. ► Flexible road models can handle unbranched and branching roads. ► Sensor data integration and road model transitions are naturally implemented in a particle filter-based framework. ► We test the method in various outdoor scenes.