高斯变分推理的无人机状态与轨迹估计方法

UAV state and trajectory estimation method based on Gaussian variational inference

针对目前的状态估计算法在面对非线性大批量状态时,存在的误差过大、算法迭代次数过多等问题,通过引入变分推断方法,提出了无人机轨迹的高斯变分推断(gaussian variational inference,GVI)精确估计方法.该方法首先通过提出基于高斯变分推断的损失函数,将状态估计问题转化为利用数据对后验进行近似的问题.然后,采用牛顿式更新以及梯度下降法的思想对损失函数、均值以及协方差矩阵进行优化迭代.使用该算法对无人机的状态以及轨迹进行估计,仿真结果表明,本算法精度较高.同时,本算法与最大后验估计(maximum a posteriori,MAP)算法相比,能够有效降低损失函数值,提高轨迹估计的精确性.

Aiming at the problems of excessive error and too many algorithm iterations in the current state estimation algorithm in the face of nonlinear mass state,by introducing the idea of variational inference,it is proposed that the Gaussian variational inference(GVI)of the UAV trajectory is accurate method of estimation.This method first converts the problem of state estimation into a problem of approximating the posterior by using data by proposing a loss function based on Gaussian variational inference.Then,the Newtonian update and gradient descent method are used to optimize the loss function,mean value and covariance matrix.Using this algorithm to estimate the state and trajectory of the UAV,the simulation results show that the algorithm has high accuracy.At the same time,compared with the maximum a posteriori estimation(MAP)algorithm,this algorithm can effectively reduce the loss function value and improve the accuracy of trajectory estimation.