Bezier-based trajectory generation for autonomous ground vehicles

This project endeavors to create an algorithm to efficiently plan trajectories for autonomous land vehicles through two distinct environments. The first environment is akin to a racecourse, consisting of straight corridors and corner sections. The course may also include obstacles which must be traversed in a slalom fashion. The second environment is a field of binary colored cones. One color indicating traversal to the left and the other indicating traversal to the right. The only difference between these two scenarios is how the course is initially constructed, otherwise there is no difference between how path planning occurs. The course is first decomposed into a series of waypoints and corridor widths of straight lines connecting successive way points. The course is then further stratified into straight cells and corner cells. The area of the corner cells is then discretized into points called "gates." Straight lines connecting the gates in each corner cell are defined. A given set of gates then defines a piecewise linear trajectory. This path is discontinuous in the first derivative at each gate point and must be smoothed with some function. B'{e}zier curves of degree 3 are used for initial smoothing of the path. This leaves a discontinuity in the second derivative which is remedied by using a 5-point B'{e}zier curve. Slalom sections are handled by defining groups of gates on alternating sides of successive obstacles within the same cell. Each cell containing a slalom Section~must now be planned through twice to account for both possible sides in traversing the first obstacle. Cones are treated by placing way points at a user defined distance from cones on the appropriate side relative to the point of approach. The final step of this project is to implement these techniques in C++ on an actual vehicle using the Willow Garage Robotic Operating System (ROS).