Max Min Std ATE w.r.t. Translation Part (m) [With Sim(3) Umeyama Alignment] 2 Median Mean RMSE 0.5 1 ATE (m) (a) Estimate = ORB-SLAM2 2 2 1 Estimate = LTVS-FreMen 2 1 021 ATE (m) 1 2 ATE (m) (c) Estimate (d) FIGURE 10. Related statistics of the ATE of our method and other visual SLAM on the testing datasets one month after the training datasets. (a) A chart of different parameters of the ATE. (b) A comparison chart of the ATE. (c) A histogram of the ATE. (d) A violin histogram of the ATE. Max: maximum; Min: minimum; Std Dev: standard deviation; RMSE: root-mean-square error w.r.t.: with respect to. TABLE 5. A comparison of the ATE and correctness in sequences with different times after the last observation. ORB-SLAM2 LTVS-FREMEN SEQUENCES One day later One week later One month later RMSE MAX CRf,z RMSE MAX LTVS-ARMA CRf,z RMSE MAX LTVS-BPF CRf,z RMSE MAX CRf,z 0.4563 0.7044 99.99% 0.4583 0.7343 99.99% 0.4325 0.7911 99.99% 0.4219 0.7038 99.99% 0.4935 1.2742 95.5% 0.4938 1.5531 95.15% 0.4537 1.3019 98.05% 0.4603 1.0018 97.76% 0.8473 2.0697 83.27% 0.5902 1.4737 91.33% 0.5849 1.5261 90.05% 0.5255 1.627 92.15% Two months later 0.8515 2.0021 81.17% 0.6311 1.8519 87.78% 0.6129 1.9085 88.87% 0.5805 1.8562 90.17% RMSE: root-mean-square error; Max: maximum. 48 IEEE ROBOTICS & AUTOMATION MAGAZINE MARCH 2023 Estimate = LTVS-BPF 0.5 1.5 1 Estimate = LTVS-ARMA 2.5 1.75 1.5 1.25 1 ORB-SLAM2 LTVS-BPF LTVS-FreMen LTVS-ARMA 1.5 2 0.75 0.5 0.25 050 100 t (s) (b) 150 200 ORB-SLAM2 LTVS-BPF LTVS-FreMen LTVS-ARMA ATE (m) ATE (m) ORB-SLAM2 LTVS-ARMA LTVS-FreMen LTVS-BPF