Collaborative Research: Negotiated Planning for Stochastic Control of Dynamical Systems

该项目侧重于开发新的计算工具和新知识，可用于帮助卫星地面操作人员管理下一代空间任务的复杂性。航天器的地面操作人员通常必须平衡多个相互冲突的目标，随着航天器任务变得越来越复杂，地面操作人员的卫星协调任务也将变得越来越复杂。然而，在为卫星设计路径时，现有的工具使运营商很难完全了解可能的权衡和回报。此外，使用自主性引导卫星沿着期望的路径可能会引入进一步的复杂性和不确定性。该项目支持以下问题的研究：在不确定环境中运行的自主系统的路径规划如何对人类、动态和适当的风险水平做出反应？创建一个数学和算法框架来实现这些目标，可能会对涉及航天器以外其他领域自动驾驶车辆的复杂任务产生更广泛的影响。该基金支持算法和理论方法的开发，使人类操作员能够在受控自动驾驶车辆的路径规划中无缝地操纵任务目标、风险和奖励。该研究方法的前提是，凸优化不仅为随机运动规划和控制提供了理论框架，而且还为任务参数的风险、回报和约束的敏感性分析提供了理论框架，这在很大程度上是由于它能够以运行时兼容的方式提供证书。PIs的重点是开发系统的方法和工具：1)在不需要专家知识的情况下，规范任务目标和约束条件；2)用户与车辆自主控制系统协商奖励参数、风险承受能力和约束条件；3)将这些能力整合到后退视界框架中，以响应任务优先级和操作员偏好的意外和动态变化。本研究的新颖之处在于将数据驱动的不确定性表征纳入随机最优控制框架；运用对偶理论对目标、风险和回报进行敏感性分析；并在运行时实现随机可达性和后退地平线框架内的优化算法，以实现实时操作员支持。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

This project focuses on the development of new computational tools and new knowledge that can be used to help ground operators of satellites manage the complexity of next generation space missions. Ground operators of spacecraft typically must balance multiple, conflicting goals, and as spacecraft missions become more complex, so will the ground operator's task of satellite coordination. However, existing tools make it difficult for operators to obtain a complete understanding of possible trade-offs and rewards when designing paths for the satellites to follow. Further, the use of autonomy to guide satellites along desired paths can introduce further complexity, as well as uncertainty. This project supports research that is motivated by the question: How can path planning for autonomous systems operating in uncertain environments, be responsive to the human, the dynamics, and appropriate levels of risk? Creation of a mathematical and algorithmic framework to accomplish these objectives could have broader impact on complex missions involving autonomous vehicles in other domains beyond spacecraft.This grant supports the development of algorithms and theoretical methods to enable the human operator to seamlessly manipulate mission objectives, risks, and rewards in path planning for controlled autonomous vehicles. The research approach is premised on the notion that convex optimization provides a theoretical framework for not only stochastic motion planning and control, but also for sensitivity analysis of the risks, rewards, and constraints, to mission parameters, in large part due to its ability to provide certificates in a run-time compatible manner. The PIs focus on the development of systematic methods and tools for 1) specification of mission objectives and constraints without the need for expert knowledge; 2) negotiation of reward parameters, risk tolerances, and constraints, between the user and the vehicle's autonomous control system; and 3) integration of these capabilities into a receding horizon framework, to enable responsiveness to unanticipated and dynamic changes to mission priorities and operator preferences. The novelty of this research is in the inclusion of data driven characterization of uncertainty into a stochastic optimal control framework; in the use of duality theory for sensitivity analysis of objectives, risks, and rewards; and in the run-time implementation of stochastic reachability and optimization algorithms within a receding horizon framework, to enable real-time operator support.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.