Ryoma expected to go to the DMZ on Saturday. The Korean government, on the other hand, had other ideas and cancelled it. Because of this, Ryoma had to find something to do. He did remember a recommendation made by one of his labmates: Deoksogung Palace, easily the most interesting of the palaces he had seen.
An oddly placed gate at Deoksogung, one of the more traditional pieces of architecture in the palace. Ryoma’s interest in the area came from the fact that Deoksogung was the residence of King Gojong, the last king of the Joseon Dynasty and the first emperor of the (short-lived) Korean Empire. Being Korea’s analogue to Japan’s Emperor Meiji (in Ryoma’s eyes at least), one of Gojong’s major goals in his rule was Korea’s modernization. This castle shows many signs of that goal, from chandeliers in otherwise traditional buildings, to more untraditional additions.
Jeonggwanheon, the king’s lounge. This was one of the first pieces of western architecture that Ryoma saw at the site. Combining romanesque columns from the west, and Korean imperial symbols and color motifs, this building exhibits an interesting blend of east and west
Seokjojeon, the least traditional of all the buildings. This building was built as the sleeping quarters of the Emperor. It was finished in 1910, the same year Japan conquered Korea. Inside the building is an art museum.
After that, Ryoma was hungry. So, he consulted the Michelin guide and went to Buchon Yukhoe, believing Yukhoe to be a specialty item he would have to pay quite a bit for regardless. In trying to head there, he wound up discovering Gwangjang market on accident, finally finding one of the bustling street markets he kept trying to look for in Korea.
A section of Gwanjang Market. After seeing places like Myeongdong completely closed up, Ryoma was pleasantly surprised to find a place like this.
A bowl of Yukhoe with sides. This is, of course, marinated raw meat.
An excavation site underneath Makercity Sewoon. Ryoma, found this along the path to the nearest subway station.
With a week left, Ryoma did not have much time to finish his work in Korea. With that, he cut his weekend short and went to work the next day
Over the last week, Ryoma ditched RTABMAP as his SLAM algorithm, and adopted ORB-SLAM2 instead. After making this switch, Ryoma finally saw success in implementing a visual SLAM algorithm. The switch was not entirely without its problems however. While the algorithm works as intended on a real camera, the sparse detail in a virtual environment presented problems for the algorithm, which works off the presence of notable features such as corners and lines. As smooth surfaces do not have such markers, the algorithm would fail at untextured portions of the environment.
Because of this, Ryoma will have to port his simulation onto an engine with higher fidelity. Unity has modules to import URDF files and communication with ROS, making it a logical choice for an alternative engine.
Ryoma also discussed the possibility of writing a project paper that utilizes his work over the last seven weeks.
However, he does have only one week left, which could make for a horrific crunch if things do not go well.
Visual SLAM algorithm implemented in Rviz, with a point cloud. The algorithm detects certain landmarks in the picture (e.g. corners, straight lines) and assigns each point a distance value through stereography. While the technology is similar to the depth camera, the algorithm takes in fewer points saving space for data collection while retaining important features in the image.
The same visual SLAM algorithm implemented in a virtual robot. While the algorithm functions the same as it did before, it is not without issues. Because the algorithm requires the presence of detail to extract its points, the algorithm cannot function well in a sparse environment which is why the algorithm cannot pick up points on the far wall or the floor. For the algorithm to be emulated properly in a virtual environment, the simulation assets should be ported to an engine with higher visual fidelity.
Now, all he has to do is to enable 3D mapping and localization using rtabmap. While this is currently eluding him, he is getting closer to getting these algorithms to work.
The final workshop session was practice for a presentation reporting on the cohort’s experiences in Korea. This was a terrible failure for Ryoma, who will have to revise his routine entirely. He is in need of a redemption.