typingcat / spatial-topology-teleoperation Goto Github PK

This feature was developed before this package was created and tested on issue #6. There were two issues and LiDAR-Thinning approach was adopted. Let me introduce the history.

There are several ways to implement nodes of graph. I can use hash or existing python modules. Hash is fast, and existing python modules have various functions. Searching and testing are needed to make a decision.

I'm using a modified Turtlebot3 Waffle from Robotis. I just call this robot waffle which is also the name of this project.

• Maximum translational velocity: 0.6m/s
• Weight: 6.2kg
• Wheel radius: 0.08m
• Turning radius: 0.25m

I'd like to test how well it extracts topology. LiDAR publish observations at 12Hz, but it is slower in the topology extraction process. The extraction area is limited to 5m x 5m.

Vscode debugger and intellisense have different interpretations of import.

I'd like to separate the graph class into a new file. However, there was a problem in importing a new file from an existing file. The followings are error messages according to the import method.

• from graph import *
  • Pylance: Import "graph" could not be resolved
  • ROS2: No module named 'graph'
• from .graph import *
  • Debugger: Attempted relative import with no known parent package

The robot should be able to create the topology from the observations to guide users in real time. No building plans or maps are given. Therefore graph merging is an essential function for topology construction. Issue #7 mentioned a valid approach to this problem. I have to embody this approach to code level.

There are 2 suggestions what the spatial structure represents. I think the first one is better.

1. Every nodes
   • Represent spatial structure directly
   • Heavy
2. Intersection nodes only
   • Has angles of its legs
   • Light

Before studying intersection double clustering(#17), modularize intersection position clustering first.

I noticed that rviz doesn't print markers on proper position. This bug can be easily reproduced.

1. Publish a static marker at (1, 0, 0) in base_footprint.
2. In rviz, make sure the marker is fixed.
3. Change frame_id to odom.
4. You can see that the marker is no longer fixed.

It needs to be solved to extend the topology graph to the global coordinate system.

Graph structure is needed to improve topology. Until now, nodes and edges are defined as lists. The list is simple but is not efficient for searching neighbors. The new graph structure is as follows:

• Undirected Graph
  • Nodes
    • ID
    • Position
    • Linked neighbor nodes
  • Functions

Note that markers are not included in the graph class. Only essential members are included such as searching, traversal.

Better visualization tool is required to do more complex experiments.

Visualization tool for this project should be able to plot graphs over time. Python module plotly supports animation tools, and it requires accumulated data. I tried accumulating data of test #13, but it's too large to fit into the list.

Test #13 showed the possibility of the clustering approach. However, the current structure trains a new clustering model every time for the last 30 samples. Even there are no intersection angle. It need to be improved.

This test is a extension of issue #6.

Previous test shows the performance of the topology extraction computation. On the other hand, the concern of this issue is whether to extract intersections well. Following is the dataset to be used for testing. My robot was moving through the hall for a minute to collect the topology. There are 5 intersections.