Approach an autonomous vessel as a single robot with Robot Operating System in virtual environment

Figures & data

Figure 1. Simulation environment (setting and communication).

Figure 2. Comparing the loading time between HTTP/1.1 and HTTP/2 (De Saxcé, Oprescu, and Chen 2015).

Figure 3. Vessel model with coordinate systems (base_link, navigation, base_scan) in ROS RVIZ, and Unity (NJORD 2021).

Figure 4. Coordinates transformation.

Figure 5. Constructed coordinates in the model.

Figure 6. Communication diagram by node while mapping (RQT).

Figure 7. Schematic diagram of the proposed autonomous navigation with object detection model.

Figure 8. Simulation environment in the unity.

Figure 9. Projected environment by Lidar in ROS RVIZ.

Figure 10. Algorithm for optimal route.

Figure 11. Algorithm for action by object detection.

Figure 12. LabelImg program for labelling and labeled text file (representing class, x position, y position, width, height).

Figure 13. YOLO v4 models – Training process (red line – mAP, blue line – Loss).

Model 1–0.0013, 64/8, 320 × 160Model 2–0.0013, 64/8, 640 × 320Model 3–0.0013, 64/16, 640 × 382Model 4–0.0013, 64/16, 640 × 352Model 5–0.00261, 64/8, 320 × 160Model 6–0.00261, 64/8, 640 × 352

Figure 13. Continued.

Figure 14. YOLO v4 – tiny models – Training process (red line – mAP, blue line – Loss).

Model 7–0.0013, 64/8, 640 × 352Model 8–0.00261, 64/8, 320 × 160Model 9–0.00261, 64/8, 640 × 352Model 10–0.00261, 128/8, 640 × 352

Table 1. Performances of YOLOv4 by Hyperparameter

Model	Learning rate	Batch-size/Subdivision	Resolution (Width x Height)	mAP (%)	Average IoU (%)	F1-score	Total detection time (sec)
Model 1	0.0013	64/8	320 x 160	91.90	66.12	0.85	3
Model 2	0.0013	64/8	640 x 320	94.87	70.89	0.89	5
Model 3	0.0013	64/16	640 x 384	93.19	71.3	0.89	5
Model 4	0.0013	64/16	640 x 352	94.28	70.63	0.88	5
Model 5	0.00261	64/8	320 x 160	91.70	68.37	0.86	2
Model 6	0.00261	64/8	640 x 352	94.57	72.39	0.89	4

Table 2. Performances of YOLOv4-tiny by Hyperparameter

Model	Learning rate	Batch-size/Subdivision	Resolution (Width x Height)	mAP (%)	Average IoU (%)	F1-score	Total detection time (sec)
Model 7	0.0013	64/8	640 x 352	55.08	69.63	0.57	2
Model 8	0.00261	64/8	320 x 160	9.71	64.53	0.13	1
Model 9	0.00261	64/8	640 x 352	56.61	72.37	0.59	2
Model 10	0.00261	128/8	640 x 352	55.16	72.05	0.60	2

Figure 15. Confidence by lateral.

Figure 16. Testing by unseen picture (red lateral: confidence), YOLOv4.

De Saxcé, H., I. Oprescu, and Y. Chen “Is HTTP/2 Really Faster than HTTP/1.1?” 2015 IEEE Conference on Computer Communications Workshops (INFOCOM WKSHPS), 2015. Hong Kong, China: IEEE, 293–299.

 Google Scholar

NJORD 2021. https://www.njordchallenge.com/the-competition/rules

 Google Scholar