Improving Tree-Based Trajectories Through Order Reduction/Expansion and Surrogate Models
Baron Johnson, Rick Lind · 2010
Sampling-based trajectory planners provide feasible but suboptimal trajectories through complicated environments with relatively low computational cost. The inclusion of simple motion and trajectory primitives in the sampling procedure ensures feasibility of the paths without significantly increasing computational cost. Complete optimal trajectory solutions subject to differential constraints typically require very high computational cost. Deterministic and probabilistic methods are presented which provide improvements upon such trajectories. An approach to reduce the trajectory length by replicating it with few intermediate configurations is shown to be effective at both reducing the complexity and length of the path. An approach using computational geometry to perform turns as close to constraining obstacles as possible is also shown to be effective at reducing the path length while maintaining feasibility. Surrogate modeling is also presented as an approach for improving waypoint location based on any cost function, not strictly path length.