Fig. 1. SLAM optimization algorithm process and framework

Fig. 2. Feature vector of point pi within a local neighborhood

Fig. 3. Likelihood field map during scan matching

Fig. 4. Point clouds and feature vectors in degraded and structured environments. (a) Degraded environment; (b) structured environment

Fig. 5. Loop closure constraint in loop detection

Fig. 6. The ninth teaching building under Baidu Maps

Fig. 7. The floor plan of the ninth teaching building

Fig. 8. Physical diagram of AGV

Fig. 9. Front-end experimental AGV motion trajectory diagrams. (a) AGV walking trajectory of Hector-SLAM algorithm; (b) AGV walking trajectory of Cartographer algorithm; (c) AGV walking trajectory of proposed algorithm

Fig. 10. Maps of the ninth teaching building obtained by different frontend algorithms. (a) Map generated by Hector-SLAM algorithm;(b) map generated by Cartographer algorithm; (c) map generated by proposed algorithm

Fig. 11. Feature point selection for long corridor environments in front-end experiments

Fig. 12. Line plots of the relative errors of the three algorithms in a long corridor environment

Fig. 13. Loop closure detection experiment maps. (a) Map generated by Cartographer algorithm; (b) map generated by proposed algorithm

Fig. 14. The global map of the ninth teaching building generated by proposed algorithm