http://www2.informatik.uni-freiburg.de/~stachnis/pdf/grisetti07tro.pdf

一: 论文提出了两个approach：

1、自适应重采样 ：(adaptive techniques )减少粒子数，重采样减小粒子损耗( particle depletion)，每一个粒子都是一个单独的地图。

2、提议分布：不仅考虑到机器人的运动，而且考虑到最近的观察。极大释然估计，scan-matching 最近的里程计信息。

二：MAPPING WITH RAO-BLACKWELLIZED PARTICLE FILTERS

            公式：p(x1:t,m | z1:t,u1:t−1) = p(m | x1:t,z1:t) · p(x1:t | z1:t,u1:t−1). 

参数说明：the map m and thetrajectory x1:t = x1,...,xt of the robot. This estimation is performed given the observations z1:t = z1,...,zt and the odometry measurements u1:t−1 = u1,...,ut−1 obtained by the mobile robot. 

该公式表明可以先估计robot的pose，再根据轨迹估算map。这个方法就叫做 Rao-Blackwellization。

根据公式 ：p(m | x1:t,z1:t)，估算map，因为  x1:t and z1:t 是已知的。

根据公式： p(x1:t | z1:t,u1:t−1) ，利用粒子滤波，估算pose。

常用的粒子滤波算法是：sampling importance resampling (SIR) filter。

三、 RBPF WITH IMPROVED PROPOSALS AND ADAPTIVE RESAMPLING

A. On the Improved Proposal Distribution （改进提议分布）

Unfortunately, in the context of SLAM a closed form of this posterior is not available in general. As a result, typical particle filter applications [3, 29] use the odometry motion model as the proposal distribution. 

B. Efficient Computation of the Improved Proposal

C. Discussion about the Improved Proposal

D. Adaptive Resampling

E. Algorithm

四. IMPLEMENTATION ISSUES

五. COMPLEXITY 

六. EXPERIMENTS

七. RELATED WORK