Reducing Complexity in Routing of Non-Standard Intersections, to Aid in Autonomous Vehicle Navigation

Autonomous vehicles must possess the capability to navigate complex intersections, which do not conform to typical models. Such intersections may have multiple roadways of different classes, highly acute angles, or unique multi-modal combinations. These may include railway grade crossings, bicycle lanes, or unique signal arrangements. Conventional navigation systems, which gather data from the surrounding area then plan a path through the collected data require faultless and complex analysis of extremely unstructured environments. The vehicle must then avoid obstacles as well as successfully navigate the intersection with extremely low tolerance for error. Computer decision making challenges can arise from this method of navigation, especially when interacting with non-autonomous vehicles. This research presents a computational method of simplifying road intersections based on pre-planned routing data to enable navigation through complex intersections with minimal instruction sets The static nature of roadways enabled detailed path planning, using a series of lines and arcs, which reduced, even the most complex intersections, into simply navigable splines. A five way, high angle intersection, including multiple railroad grade crossings and non-standard markings, was replicated for this small-scale evaluation. The prototype autonomous vehicle then navigated the intersection, in a typical routing permutation, without the aid of external sensors. This method reduces the risk associated with navigational miscues, enabling a robust network enabled autonomous navigation model and could suggest a higher survivability in the case of sensors failure. The results of this research provide a robust method for intersection navigation, which does not require standardized marking or traffic management cues beyond vehicle localization and pre-planned route spline data. By tracking the vehicle's translation and attitude, path corrections maintain tracking on the virtual spline. By measuring the tangential and radial accelerations, the test platform demonstrated smooth navigation of one permutation of the intersection, with no external sensor input. Spiro-circular modification of the path reduced the episodic jerk present in simple circular routing definitions This system enables safer navigation of complex environments, while the vehicle's environmental and obstacle sensors may be used to provide episodic modification to planned routes in execution, rather than be relied upon for primary navigation.