The field of quadrotor motion planning has experienced significant advancements over the last decade. Most successful approaches rely on two stages: a front-end that determines the best path by incorporating geometric (and in some cases kinematic or input) constraints, that effectively specify the homotopy class of the trajectory; and a back-end that optimizes the path with a suitable objective function, constrained by the robot's dynamics as well as state/input constraints. However, there is no systematic approach or design guidelines to design both the front and the back ends for a wide range of environments, and no literature evaluates the performance of the trajectory planning algorithm with varying degrees of environment complexity. In this paper, we propose a modular approach to designing the software planning stack and offer a parameterized set of environments to systematically evaluate the performance of two-stage planners. Our parametrized environments enable us to access different front and back-end planners as a function of environmental clutter and complexity. We use simulation and experimental results to demonstrate the performance of selected planning algorithms across a range of environments. Finally, we open source the planning/evaluation stack and parameterized environments to facilitate more in-depth studies of quadrotor motion planning, available at https://github.com/KumarRobotics/kr_mp_design

翻译：过去十年间，四旋翼飞行器运动规划领域取得了显著进展。多数成功方法依赖两个阶段：前端通过融入几何约束（部分情况下包含运动学或输入约束）确定最优路径，有效指定轨迹的同伦类；后端则基于合适的代价函数对路径进行优化，并受限于机器人动力学及状态/输入约束。然而，目前缺乏针对多样化环境设计前后端模块的系统化方法或设计准则，现有文献也未评估轨迹规划算法在不同环境复杂度下的性能。本文提出一种模块化软件规划栈设计方法，并提供参数化的环境集合以系统评估两阶段规划器的性能。通过参数化环境，我们能够根据不同环境杂乱度与复杂度评估不同前端及后端规划器。利用仿真与实验数据，本文展示了所选规划算法在多种环境中的性能表现。最后，我们开源了规划/评估栈及参数化环境集合（https://github.com/KumarRobotics/kr_mp_design），以促进四旋翼运动规划的深入研究。