轮廓搜索提议密度粒子滤波与弹性地形基准导航
Contours-Seeking Proposal Density Particle Filter and Resilient Terrain-Referenced Navigation
浏览论文内容
中文总结 AI 辅助
针对粒子滤波退化问题，提出含高斯混合随机强迫机制的轮廓搜索采样策略，经实验验证可降低权重方差、提升有效样本量，在恶劣场景下仍能可靠运行。
中文摘要 AI 辅助
辅助导航系统对飞行器的可靠运行至关重要，尤其是在地形基准导航这类自主框架中。然而，多模态似然、高度非线性地形高程以及未知预测偏差等挑战会导致后验分布高度多模态且可预测性差，进而引发粒子滤波退化。本研究针对粒子滤波方法的数值不稳定性与退化问题，提出了一种适配该问题的采样策略。该方法引入高斯混合随机强迫机制，沿地形坡度推动粒子，对抗向最可能地形轮廓偏移的趋势；每个混合分量对应一个似然模式，提升对未建模地形特征的适应性。为进一步提升有效性，辅助采样有选择性地将该混合采样应用于大概率粒子，得到退化程度更低、权重分布更均匀的粒子集。数值实验验证了所提方法在降低权重方差、提升有效样本量方面的有效性；此外，该方法在恶劣场景（如严重未知预测偏差、多模态测量噪声）下表现出强弹性，可确保长期可靠的粒子滤波。
英文摘要
Auxiliary navigation systems are essential for the robust operation of aerial vehicles, particularly in self-contained frameworks like terrain-referenced navigation. However, challenges such as multimodal likelihoods, highly nonlinear terrain elevations, and unknown prediction biases result in highly multimodal and less predictable posterior distributions, leading to particle filter degeneration. This study addresses the numerical instability and degeneration of the particle filter approach by proposing a sampling strategy tailored to this problem. The approach introduces a Gaussian mixture random forcing mechanism, which nudges particles along terrain slopes and against biases towards the most probable terrain contours. Each mixture is associated with a mode of likelihood, enhancing adaptability to unmodeled terrain features. To further improve effectiveness, auxiliary sampling selectively applies this mixture sampling to probable particles, yielding a less degenerate and evenly weighted particle set. Numerical experiments demonstrate the effectiveness of the proposed method in reducing weight variance, improving effective sample size. In addition, the approach exhibits strong resilience under deteriorating scenarios, such as severe unknown prediction bias and multimodal measurement noise, ensuring long-term reliable particle filtering.
