Robust Path Planning and Feedback Design Under Stochastic Uncertainty
Robust Path Planning and Feedback Design Under Stochastic Uncertainty
复制标题
随机不确定性下的鲁棒路径规划与反馈设计
DOI：
10.2514/6.2008-6304
复制
发表时间：
2008
影响因子：
--
通讯作者：
L. Blackmore
中科院分区：
文献类型：
--
作者：
L. Blackmore
自动驾驶车辆需要最优路径规划算法来实现任务目标，同时避开障碍物并对不确定性具有健壮性。不确定性由外部扰动、建模误差和传感器噪声引起，这些不确定性可以通过随机模型来表征。以前的工作通过使用机会约束的概念削弱了随机环境中的稳健性的概念。这就要求违反任务约束的概率可能小于规定值。提出了一种带反馈设计的机会约束最优路径规划的新方法。该方法优化了待跟踪的参考轨迹和用于抑制不确定性的反馈控制器。我们的方法推广了最近基于凸优化的约束控制综合的结果，以解决非凸约束的控制问题。这种扩展对于路径规划问题是必不可少的，因为路径规划问题本身就具有非凸的避障约束。与以往的机会受限路径规划方法不同，新方法在优化参考轨迹的同时优化了反馈增益。其关键思想是将不考虑不确定性的快速非凸求解器与仅适用于凸可行域的现有稳健方法相结合。通过在稳健和非稳健解之间交替，新算法确保收敛到全局最优解。将该方法应用于一架无人机，仿真结果表明了该方法的有效性。
Autonomous vehicles require optimal path planning algorithms to achieve mission goals while avoiding obstacles and being robust to uncertainties. The uncertainties arise from exogenous disturbances, modeling errors, and sensor noise, which can be characterized via stochastic models. Previous work dened a notion of robustness in a stochastic setting by using the concept of chance constraints. This requires that mission constraint violation can occur with a probability less than a prescribed value. In this paper we describe a novel method for optimal chance constrained path planning with feedback design. The approach optimizes both the reference trajectory to be followed and the feedback controller used to reject uncertainty. Our method extends recent results in constrained control synthesis based on convex optimization to solve control problems with nonconvex constraints. This extension is essential for path planning problems, which inherently have nonconvex obstacle avoidance constraints. Unlike previous approaches to chance constrained path planning, the new approach optimizes the feedback gain as well as the reference trajectory. The key idea is to couple a fast, nonconvex solver that does not take into account uncertainty, with existing robust approaches that apply only to convex feasible regions. By alternating between robust and nonrobust solutions, the new algorithm guarantees convergence to a global optimum. We apply the new method to an unmanned aircraft and show simulation results that demonstrate the ecacy of the approach.