A Study Of SLAM Back-end Optimization Algorithm Based On G2o

With the development of mobile robot, more and more people are making a research on robot. SLAM(simultaneous localization and mapping) is the basis of making a mobile robot truly autonomous. Currently solutions to the problem of SLAM are classified into two types: filtering and smoothing, and smoothing is also called graph-based SLAM. Graph-based SLAM is divided into two parts: front-end and back-end. The pose graph is constructed by the front-end, and the back-end in turn optimizes the pose graph. Since the place recognition algorithm in the front-end produces the false-positive loop closure, which makes the back-end solvers converge towards a defective solution. Therefore, a proposed robust back-end optimization algorithm is the hot research.This paper studies the currently typical back-end optimization algorithms: SC(Switchable Constraints), RRR(Realizing, Reversing. Recovering), MM(Maxmixture) and DCS(Dynamic Covariance Scaling). By using standard datasets testing RRR and MM algorithms, we found that they failed when optimize some datasets. Since SC algorithm increases the scale of the problem, so that computational complexity is increased. DCS algorithm is proposed on the basis of SC algorithm, but for some datasets it fails. The paper proposes an improved DCS(DCS1) algorithm by analyzing the objective function of SC algorithm. Meanwhile, this paper makes a detailed analysis and comparison on SC, RRR, MM and DCS1 algorithms, and analyzes the controllable parameters of these algorithms.This paper describes all the standard datasets and makes a series of tests using these datasets. The experimental results compared to DCS algorithm verify the correctness of DCS1 algorithm and the necessity of improving. Then, this paper gives a comparison experimental results of SC algorithm and the improved DCS algorithm, which verify the improved DCS algorithm converges faster than the SC algorithm. Finally, this paper tests the RRR algorithm and the MM algorithm. and analyzes the advantage and disadvantage of these two algorithm.