基于状态依赖分解的保安全近似可达性
Guaranteed-Safe Approximate Reachability via State Dependency-Based Decomposition
摘要:
Hamilton Jacobi(HJ)可达性是一种广泛应用于机器人安全性分析的形式化验证工具。给定一个不安全状态的目标集,如果动力系统避开后向可达管(BRT),则保证该系统在最坏情况下不进入目标。然而,计算BRT遭受相对于状态维度的指数计算时间和空间复杂度。以前,系统分解和投影技术已被调查,但适用于更广泛的一类动态和保守程度之间的权衡一直具有挑战性。本文提出了一种状态依赖图来表示系统的动态特性,并对整个系统进行了分解,其中每个子系统中只包含依赖状态,“缺失”状态被视为有界扰动。因此,对于机器人技术中的各种动态,BRT可以快速地近似于低维链式子系统,并保持其安全性。我们证明了我们的方法与数值实验的4D四重积分器,和6D自行车,一个重要的汽车模型,以前是棘手的。
英文摘要:
Hamilton Jacobi (HJ) Reachability is a formal verification tool widely used in robotic safety analysis. Given a target set as unsafe states, a dynamical system is guaranteed not to enter the target under the worst-case disturbance if it avoids the Backward Reachable Tube (BRT). However, computing BRTs suffers from exponential computational time and space complexity with respect to the state dimension. Previously, system decomposition and projection techniques have been investigated, but the trade off between applicability to a wider class of dynamics and degree of conservatism has been challenging. In this paper, we propose a State Dependency Graph to represent the system dynamics, and decompose the full system where only dependent states are included in each subsystem, and "missing" states are treated as bounded disturbance. Thus for a large variety of dynamics in robotics, BRTs can be quickly approximated in lower-dimensional chained subsystems with the guaranteed-safety property preserved. We demonstrate our method with numerical experiments on the 4D Quadruple Integrator, and the 6D Bicycle, an important car model that was formerly intractable.
机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。