TRIPODS+X:RES: Safe Imitation Learning for Robotics

学习算法现在广泛应用于机器人系统中。然而，对预测可接受性具有高概率理论保证的安全学习程序的研究远远不够，特别是对于使用专家收集的数据进行训练并按顺序做出决策的机器人系统。pi将汇集统计学、机器学习和数学优化方面的思想和技术，为与人类互动的几类现代机器人设计下一代模仿学习方法，并提供可证明的安全保证。本项目旨在：(1)开发安全模仿学习的新配方；(2)设计具有理论安全性保证的快速学习算法；(3)探索有益人机交互的信任建立过程。研究方法将在几个机器人问题领域进行评估，包括机器人操作和轮式移动机器人导航。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。

Learning algorithms are now pervasively deployed in robotic systems. However, safe learning procedures with high-probability theoretical guarantees on the acceptability of the predictions have been far less studied, especially for robotic systems trained with data collected from experts and making decisions sequentially. The PIs shall bring together ideas and techniques from statistics, machine learning, and mathematical optimization, to design the next generation of imitation learning approaches with provable safety guarantees for several classes of modern robots that interact with humans. The project aims to: (1) develop new formulations of safe imitation learning; (2) design fast learning algorithms with theoretical guarantees on safety; (3) explore trust-building processes for beneficial human-machine interaction. The research approaches will be evaluated in several robotic problem domains, including robotic manipulation and wheeled mobile robot navigation.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.