基于图优化的同时定位与建图(SLAM)系统中含有大噪声的回环边,可能严重阻碍优化器迅速收敛到最优解,显著降低定位精确性和地图一致性.因此,针对大噪声回环边的优化算法的鲁棒性至关重要.引入K-means聚类思想,对回环边残差值进行分类,进而建立了一种新的残差阈值模型,自适应调整回环边在优化时的权重,减少回环边对优化的影响;然后,基于迭代重加权最小二乘的思想形成了 RW-RLSPGO算法(residual weighted enhancement for recursive least squares pose graph optimization algorithm,RW-RLSPGO);最后,在模拟和真实的PGO数据集上进行蒙特卡罗实验.实验结果表明,RW-RLSPGO算法在准确性和鲁棒性方面都取得了显著的提高,验证了其在大噪声环境下的有效性.
Pose graph optimization algorithm based on loop-closure edges residual focusing weight model
In graph-based SLAM systems,loop-closure edges with large noise may severely impede the optimizer from rapidly converging to the optimal solution,leading to a noticeable decrease in localization accuracy and map consistency.Therefore,the objective of this paper was to investigate robust methods for handling loop-closure edges,which was crucial for optimization algorithms in the presence of large noise.Toward this aim,this paper introduced a new concept of K-means clustering to clas-sify the residual values of loop-closure edges,thereby established a new residual threshold model.This model adaptively adjus-ted the weights of loop-closure edges during optimization to reduce their impact on the optimization process.Subsequently,the formulation of the residual weighted enhancement for recursive least squares pose graph optimization algorithm(RW-RLSPGO)was based on the iterative reweighted least squares principle.Finally,it conducted Monte Carlo experiments on both simulated and real pose graph optimization(PGO)datasets.The experimental results demonstrate a significant improvement in both accuracy and robustness with the RW-RLSPGO algorithm,validating its effectiveness in high-noise environments.
simultaneous localization and mapping(SLAM)pose graph optimizationloop-closure edgelarge noiseclus-tering
万方数据
维普
