Software and Information Systems Engineering, Ben-Gurion University of the Negev(本·古里安大学软件与信息系统工程系);Department of Computer Science and Engineering, Washington University(华盛顿大学计算机科学与工程系)
Argaman Mordoch, Shahaf S. Shperberg, Roni Stern, Berndan Juba
A significant challenge in applying planning technology to real-world problems lies in obtaining a planning model that accurately represents the problem's dynamics. Obtaining a planning model is even more challenging in mission-critical domains, where a trial-and-error approach to learning how to act is not an option. In such domains, the action model used to generate plans must be safe, in the sense that plans generated with it must be applicable and achieve their goals. % Learning safe action models for planning has been mostly explored for domains in which states are sufficiently described with Boolean variables. % In this work, we go beyond this limitation and propose the Numeric Safe Action Models Learning (N-SAM) algorithm. In this work, we present N-SAM, an action model learning algorithm capable of learning safe numeric preconditions and effects. We prove that N-SAM runs in linear time in the number of observations and, under certain conditions, is guaranteed to return safe action models. However, to preserve this safety guarantee, N-SAM must observe a substantial number of examples for each action before including it in the learned model. We address this limitation of N-SAM and propose N-SAM*, an extension to the N-SAM algorithm that always returns an action model where every observed action is applicable at least in some states, even if it was observed only once. N-SAM* does so without compromising the safety of the returned action model. We prove that N-SAM* is optimal in terms of sample complexity compared to any other algorithm that guarantees safety. N-SAM and N-SAM* are evaluated over an extensive benchmark of numeric planning domains, and their performance is compared to a state-of-the-art numeric action model learning algorithm. We also provide a discussion on the impact of numerical accuracy on the learning process.
SHARP: Shielding-Aware Robust Planning for Safe and Efficient Human-Robot Interaction
Princeton University(普林斯顿大学)
在人机交互(HRI)设置中共同实现安全性和效率是一个具有挑战性的问题,因为机器人的规划目标可能与人类自己的意图和期望不一致。最近的方法,确保安全的机器人操作在不确定的环境中,通过监督控制方案,有时被称为“屏蔽”,它覆盖机器人的标称计划与安全回退策略时,一个安全的关键事件是迫在眉睫的。这些反应性的“最后手段”策略(通常以积极的紧急机动的形式)专注于保护安全而不考虑效率;当名义计划者不知道可能的安全覆盖时,屏蔽可以比必要的更频繁地激活,导致性能下降。在这项工作中,我们提出了一个新的屏蔽为基础的规划方法,允许机器人规划效率明确占未来可能的屏蔽事件。利用贝叶斯人体运动预测的最新工作,由此产生的机器人政策积极平衡标称性能与低概率人类行为引发的高成本紧急演习的风险。我们形式化屏蔽感知鲁棒规划(SHARP)作为一个随机最优控制问题,并提出了一个计算效率高的框架,在运行时找到易于处理的近似解。我们的方法在模拟驾驶示例中的性能优于屏蔽不可知的运动规划基线(配备相同的人类意图推理方案),这些示例中的人类轨迹来自最近发布的Waymo Open Motion Dataset。