Formal Synthesis of Safety Controllers for Unknown Systems Using Gaussian Process Transfer Learning
Formal Synthesis of Safety Controllers for Unknown Systems Using Gaussian Process Transfer Learning
复制标题
DOI：
10.1109/lcsys.2023.3341548
复制
发表时间：
2023
影响因子：
3
通讯作者：
A. Awan;Majid Zamani
中科院分区：
文献类型：
--
作者：
A. Awan;Majid Zamani
在这封信中，我们提出了一种数据驱动的方法，使用高斯过程（GP）迁移学习来合成未知非线性控制系统的安全控制器。我们的方法包括两个步骤。第一步涉及使用从系统采样的数据学习 GP 模型。我们的方法允许利用先前学习的相关系统（称为源系统）的 GP 模型（例如，部署在稍微不同的环境条件下的机器人）来学习当前系统（称为目标系统）的 GP 模型。在目标系统的数据收集成本高昂或耗时的情况下，这是必需的。第二步，我们根据学习到的 GP 模型计算控制障碍函数以及相应的控制器。此外，我们还量化了配备综合控制器的目标系统的安全满足概率的下限。我们通过将其应用于喷气发动机案例研究来证明所提出方法的有效性。
In this letter, we propose a data-driven approach for synthesizing safety controllers for unknown nonlinear control systems using Gaussian Process (GP) transfer learning. Our approach involves two steps. The first step involves learning a GP model using data sampled from the system. Our method allows for leveraging a previously learned GP model of a related system, known as the source system (e.g., robot deployed in slightly different environmental conditions), to learn a GP model for the system at hand, known as the target system. This is required in situations where data collection for the target system is expensive or time consuming. In the second step, we compute a control barrier function together with a corresponding controller based on the learned GP model. In addition, we quantify the lower bound on the probability of safety satisfaction for the target system equipped with the synthesized controller. We demonstrate the effectiveness of the proposed approach by applying it to a jet engine case study.