标题： 航天器在近距离操作中使用隐式神经表示对几何复杂目标的安全鲁棒控制 显示英文标题

标题： Spacecraft Safe Robust Control Using Implicit Neural Representation for Geometrically Complex Targets in Proximity Operations

摘要： 本研究解决了确保航天器近距离操作安全的挑战，重点是在存在扰动的情况下，追踪航天器与复杂几何目标航天器之间的避撞问题。 为了在这种情况下确保安全，提出了一种安全鲁棒控制框架，该框架利用了隐式神经表示。 为了处理任意目标几何形状而无需显式建模，通过增强的隐式几何正则化方法从点云数据中学习了一个神经符号距离函数（SDF），该方法结合了过近似策略以创建保守的安全优先边界。 目标的表面由学习到的神经SDF的零水平集隐式定义，而其值和梯度为安全控制器设计提供了关键信息。 这种神经SDF表示构成了一个两层分层安全鲁棒控制框架：安全速度生成层和安全鲁棒控制器层。 在第一层中，通过显式结合下近似误差界，制定了一个二阶锥规划问题，以生成保证安全的参考速度。 此外，引入了一个循环不等式，以缓解控制屏障函数（CBF）方法中常见的局部最小值问题。 第二层包括一个集成的扰动观测器和一个平滑安全滤波器，显式补偿估计误差，增强了对外部扰动的鲁棒性。 大量的数值仿真和蒙特卡洛分析验证了所提出的框架，表明与传统CBF方法相比，显著提高了安全裕度并避免了局部最小值。 显示英文摘要

摘要： This study addresses the challenge of ensuring safe spacecraft proximity operations, focusing on collision avoidance between a chaser spacecraft and a complex-geometry target spacecraft under disturbances. To ensure safety in such scenarios, a safe robust control framework is proposed that leverages implicit neural representations. To handle arbitrary target geometries without explicit modeling, a neural signed distance function (SDF) is learned from point cloud data via a enhanced implicit geometric regularization method, which incorporates an over-apporximation strategy to create a conservative, safety-prioritized boundary. The target's surface is implicitly defined by the zero-level set of the learned neural SDF, while the values and gradients provide critical information for safety controller design. This neural SDF representation underpins a two-layer hierarchcial safe robust control framework: a safe velocity generation layer and a safe robust controller layer. In the first layer, a second-order cone program is formulated to generate safety-guaranteed reference velocity by explicitly incorporating the under-approximation error bound. Furthermore, a circulation inequality is introduced to mitigate the local minimum issues commonly encountered in control barrier function (CBF) methods. The second layer features an integrated disturbance observer and a smooth safety filter explicitly compensating for estimation error, bolstering robustness to external disturbances. Extensive numerical simulations and Monte Carlo analysis validate the proposed framework, demonstrating significantly improved safety margins and avoidance of local minima compared to conventional CBF approaches.