Fig. 1. Overall framework of our algorithm

Fig. 2. Principle of laser point straight line fitting

Fig. 3. Straight line fitting. (a) Laser hit point; (b) straight line fitting

Fig. 4. Recognition of convex and concave corners. (a) Schematic of distinguishing convex and concave corners; (b) recognition result of convex and concave corners

Fig. 5. Semantic segmentation results. (a) Robot perspective; (b) semantic segmentation; (c) target detection

Fig. 6. Schematic of wall corner categories and sensor

Fig. 7. Wall corner category judgment under non-overlapping azimuth. (a)(c) Left limit view; (b)(d) right limit view

Fig. 8. Wall corner category judgment under overlapping azimuth. (a) Azimuth is 325°；(b) azimuth is 0°；(c) azimuth is 35°

Fig. 9. Wall corner directional judgment. (a)(e) Robot position; (b)(f) robot perspective; (c)(g) laser fitting line; (d)(h) wall corner category

Fig. 10. Joint calibration of camera and lidar

Fig. 11. Cabinet semantics obtained by pure semantic segmentation. (a) Robot perspective; (b) semantic segmentation result; (c) point cloud coordinate mapping result

Fig. 12. Cabinet semantics based on laser and vision. (a) Convex and concave wall corner recognition; (b) modified cabinet semantics; (c) camera depth value; (d) laser lidar depth value

Fig. 13. Overlap of semantic mapping results. (a)(d) Robot perspective; (b)(e) wall corner semantic mapping; (c)(f) object semantic mapping

Fig. 14. Convex and concave wall corner recognition in simulation environment. (a) Simulation environment; (b)－(d) convex and concave wall corner recognition

Fig. 15. Recognition of convex and concave corners. (a) Grid map; (b) concave and convex wall corner categories; (c) fused wall corner map

Fig. 16. Mobile robot platform and real environment. (a) Mobile robot platform; (b) real environment

Fig. 17. Recognition of convex and concave corners. (a) Grid map; (b) convex and concave wall corner categories; (c) four types of concave corners; (d) fused wall corner map

Fig. 18. Semantic map. (a) Object semantic map; (b) object semantic grid map; (c) final semantic grid map

Fig. 19. Real environment and semantic map. (a) Real environment; (b) grid map; (c) convex and concave wall corner categories; (d) four types of concave corners; (e) object semantic map; (f) final semantic grid map

Fig. 20. Environment map. (a) Raster map; (b) semantic map

Fig. 21. Particle convergence process of original AMCL algorithm

Fig. 22. Particle convergence process in localization using semantic map

Fig. 23. Schematics of robot pre-location based on semantic information. (a) Robot perspective; (b) wall corner category; (c) position of wall corner and robot; (d) particle distribution

Fig. 24. Particle convergence rate comparison