RRT-Based Path Planning Considering Initial and Final Pose under Curvature Constraints for Nonholonomic Wheeled Robot
RRT-Based Path Planning Considering Initial and Final Pose under Curvature Constraints for Nonholonomic Wheeled Robot
复制标题
DOI：
10.1109/iecon.2019.8927351
复制
发表时间：
2019-10
期刊：
影响因子：
--
通讯作者：
Mamoru Sobue;H. Fujimoto
中科院分区：
文献类型：
--
作者：
Mamoru Sobue;H. Fujimoto
快速探索随机树（RRT）算法是路径规划中常用的算法，因为它能有效地解决单查询问题并具有概率完备性。它的改进，如RRT*和Informed RRT*，将RRT扩展到当采样次数趋近于无穷时渐近地找到最优解。然而，这些算法没有考虑机器人的姿态和运动学约束，因此对于给定初始姿态和期望最终姿态的非完整机器人，其结果是不可行的。针对非完整路径规划问题，提出了另一种修正的RRT*，该RRT*不仅具有运动学约束，而且具有初始位姿约束和最终位姿约束。该方法将搜索树构建为一个有向图，其中每个节点保留位置（x, y）加上机器人姿态$\theta$，并使用一段曲面曲线和直线来连接和评估两个节点之间的代价。此外，该方法通过在初始位姿和最终位姿上分别构建两棵树并进行双向搜索，找到包含横切点的路径。实验表明，我们的方法计算出比RRT*更平滑、更可行的路径，并满足给定的曲率约束。
Rapidly-exploring random trees(RRT) are popular algorithms in path planning, because they provide efficient solutions to singe-query problems and possess probabilistic completeness. Its modifications, such as RRT* and Informed RRT*, extend RRT to asymptotically find optimal solutions as the number of sampling approaches infinity. These algorithms, however, give no considerations to robot's poses and kinematic constraints, and therefore their results can be unfeasible for a nonholonomic robot with given initial pose and desired final pose. In this paper, we present another modification of RRT* for nonholonomic path-planning with not only kinematic constraints but also initial and final pose constraints. The proposed method constructs the search tree as a directed graph of which each node retains the position (x, y) plus the robot pose $\theta$, and a segment of clothoid curve and line is used for connecting and evaluating the cost between two nodes. Furthermore, by building two trees from both initial and final poses and executing bidirectional search, this method can find a path containing crosscut point. We experimentally show that our approach calculates smoother, more feasible paths than RRT* and satisfy the given constraints on curvature.