Integrate with SOM

[kv13arm]

No description provided.

[MatthewMunks]

• Add interface with pymongo.
• Add new objects to the system.
• Update objects previously added into the system.
• Setup SOM and orion_actions definitions.
• Test system.
• Setup regression tests (maybe?)
• Work out the publishing of tfs. (Might just be done through task level planning, but this will represent a break from what has been done up to this point.)
• MarkerArray interface into Rviz?

[MatthewMunks]

ori-orion/orion_semantic_map@838c602
^^^ Beginning to create a class interface. The idea will be to copy this over once complete. (Want to copy over at the last moment so that updates to the MemoryManager class get carried over.)

[MatthewMunks]

ori-orion/orion_semantic_map@b9e5174
Getting the update function in.

We have object tracking, but what if that object is out of view for a little bit. Take the example of a bottle on the table. The robot turns around, and then looks again at the bottle a second time. Is the consistency across this gap covered by perception, or is it something for SOM to deal with? (Can be easily done, I just need to know!)

[MatthewMunks]

I think I'll create a config file for global params in orion_hsr_bringup so that these remain constant throughout all tasks.

[MatthewMunks]

The other question is as to the publishing of tfs. I've noticed a slight lag on the location of the tfs being published by the old recognition system on rviz. They are being published slightly ahead of the motion of the head itself (which is interesting). Whether we want the tfs being published as per the old system (maybe under slightly different names?), is something to work out I think.

[MatthewMunks]

Add an entry to som for tf_id

[MatthewMunks]

Current branches for other packages that are related to this:

• orion_semantic_map : MatthewMunks-IntegrationIntoPerception
• orion_actions: Yolov8
• orion_task_level_planning: Yolov8

[MatthewMunks]

For the tfs, can ' ' -> '_'?