Maximization of Traversal Speeds for Off-Road Mobility by Modification of Point-to-Point Paths

Open Access
- Author:
- Gruning, Veronica
- Graduate Program:
- Mechanical Engineering
- Degree:
- Master of Science
- Document Type:
- Master Thesis
- Date of Defense:
- September 03, 2020
- Committee Members:
- Sean N Brennan, Thesis Advisor/Co-Advisor
Karen Ann Thole, Program Head/Chair
Karl Martin Reichard, Thesis Advisor/Co-Advisor
Henry Joseph Sommer, III, Committee Member - Keywords:
- robotics
path planning - Abstract:
- This thesis seeks to answer the question: can a vehicle's traversal time be rapidly determined along an obstacle-prone pathway with a challenging driving surface? Current techniques in the literature that investigate this problem often fall at two extremes: 1) meticulously accurate, but computationally expensive, algorithmic techniques that use soil contact models to predict how the tires or treads of a vehicle interact with the ground surface; or 2) simplistic path planning methods that examine point-to-point path plans, but that use algorithms that ignore soil, vehicle, or slope effects with significant loss in fidelity. This thesis considers an intermediate approach that still accounts for soil/vehicle limits, yet seeks, as much as possible, to utilize the computationally efficient techniques of point-to-point path planning. The method used in this work begins with point-to-point paths generated by visibility graphs, which are calculated from the well-known A-star algorithm. These paths are then rounded to account for friction-limited turning radii around each obstacle and further optimized by varying the pivot center of each arc. The resulting calculations, while not as accurate as soil interaction modeling, enable the rapid calculation via Monte Carlo methods of relationships between uncertainty in map properties and path cost, where path cost is measured in traversal time. This thesis also compares the time-cost relationships for off-road traversals to the distance-cost relationships, perhaps the first study to do so. The risks of this approach are that the solutions discovered are local optima guided by point-to-point calculations, and, thus, global optima can be missed. However, this approach enables the rapid generation of generalizable predictions for traversal times through maps, as long as obstacle characteristics (e.g., spacing, size, etc) and friction limits (i.e., due to soil, slope, etc.) are known. Further, the resulting curves enable instant estimation of traversal times through similar maps. The results presented in this work demonstrate the development from single obstacle calculations to double obstacle calculations, as well as from complex map optimizations to cost curve analyses. Comparisons to distance-optimal trajectories show that the method, in general, produces time-optimized path plans for large (i.e., kilometer-scale) and densely populated maps extremely quickly (i.e., tens of milliseconds on a typical PC), with traversal times that are a small fraction of point-to-point path planning.