Combining Cost and Reliability for Rough Terrain Navigation

Abstract

This paper presents an improved method for planetary rover path planning in very rough terrain, based on the particle-based Rapidly-exploring Random Tree (pRRT) algorithm. It inherits the benefits of pRRT, an improvement over the conventional RRT algorithm that explicitly considers uncertainty in sensing, modeling, and actuation by treating each addition to the tree as a stochastic process. Although pRRT is well-suited to planning under uncertainty, it has limitations in minimizing the cost of path plans. Our approach addresses these limitations by considering the relevant cost functions explicitly. Such cost functions depend on the application and can include time or distance of traversal, and energy consumption of the rover. The paper demonstrates the planner performance using a specific cost function defined in terms of the energy expenditure. The improved pRRT algorithm has been experimentally validated in simulation, and it has been shown to produce lower-cost plans than the standard pRRT algorithm. The proposed approach is likely to benefit the present and future space missions as an onboard motion planner and as a ground-based tool for plan validation.