ros-llm-integration

[Kindred-Yi]

Updates:
add dexterity-interface repo to primitives, which we can use to interact with llm, once we get its response, the commands that given by llm will be saved in output.txt in this repo. Then when we launch llm-control.launch, it will automatically read output.txt and parse it to a series of different panda primitives that we already implemented, and execute these commands.
Problems:

1. llm lack understanding of each panda primitives, for example if we ask LLM to generate commands for picking the kettle, it will output move first then pick, but out pick implementation already include the "move" part. Solution: maybe give llm more prompts of panda's current capability
2. Consider include all the system in one repo and use one docker(chat.py, panda-primitives, panda-primitives-control)