The Worldwide Journal of Robotics Analysis, Forward of Print.
Movement planning below uncertainty is crucial for dependable robotic operation. Regardless of substantial advances over the previous decade, the issue stays troublesome for programs with advanced dynamics. Most state-of-the-art strategies carry out search that depends on a lot of ahead simulations. For programs with advanced dynamics, this typically requires expensive numerical integrations, which considerably slows down the planning course of. Linearization-based strategies have been proposed that may alleviate the above downside. Nevertheless, it’s not clear how linearization impacts the standard of the generated movement technique, and when such simplifications are admissible. To reply these questions, we suggest a non-linearity measure, referred to as Statistical-distance-based Non-linearity Measure (SNM), that may establish the place linearization is helpful and the place it ought to be prevented. We present that when the issue is framed because the Partially Observable Markov Resolution Course of, the worth distinction between the optimum technique for the unique mannequin and the linearized mannequin might be upper-bounded by a perform linear in SNM. Comparisons with an current measure on varied situations point out that SNM is extra appropriate in estimating the effectiveness of linearization-based solvers. To check the applicability of SNM in movement planning, we suggest a easy on-line planner that makes use of SNM as a heuristic to change between a common and a linearization-based solver. Outcomes on a car-like robotic with second order dynamics and 4-DOFs and 7-DOFs torque-controlled manipulators point out that SNM can appropriately resolve if and when a linearization-based solver ought to be used.
发布者：Marcus Hoerger，转转请注明出处：https://robotalks.cn/non-linearity-measure-for-pomdp-based-motion-planning/
