Behavior Tree Generation

Before you start

• Make sure you have the authentication to the private repos behavior_tree and behavior_tree_msgs.
• Clone these repos in same workspace:

cd ${YOUR_WORKSPACE}/catkin_ws/src

git clone [email protected]:ARG-NCTU/behavior_tree.git

git clone [email protected]:ARG-NCTU/behavior_tree_msgs.git

• And also clone this repo:

git clone [email protected]:ARG-NCTU/behavior_tree_generation.git

• Build before you run it.

cd ${YOUR_WORKSPACE}/catkin_ws

cakin_make

or

catkin_build -w ${YOUR_WORKSPACE}/catkin_ws

Params

• There are several params in behavior_tree_generation/launch/behavior_tree_generation.launch:

  • <api_key>: An unique code that identifies your requests to the open API.
  • <trigger_rosparam>(default: /LLM_generation_finished): This param define the name of a specified rosparam to trigger behavior tree node if generation is finished.
  • <model_name>(default: text-davinci-003): The large language model you what to choose.(You can see the available models here)

• And behavior_tree/launch/behavior_tree.launch also needs the follwing params:

  • <wait_for_generation>(default: false): A special param, define whether behavior need to wait tree config provided by LLM-generation.
  • <trigger_rosparam>(default: /LLM_generation_finished): This param define the name of a specified rosparam to trigger behavior tree node if param <wait_for_generation> is True.
  • :The config file for behavior tree node to read.

How to run

• First, start behavior_tree_generation.launch:

roslaunch behavior_tree_generation behavior_tree_generation.launch api_key:=${YOUR_API_KEY} trigger_rosparam:="/LLM_generation_finished" model_name:="text-davinci-003"

• Second, start behavior_tree.launch:

roslaunch behavior_tree behavior_tree.launch wait_for_generation:="true" trigger_rosparam:="/LLM_generation_finished" tree:="drone_latching/LLM_generated.tree"

• Third, start behavior_tree_rqt.launch:

roslaunch behavior_tree behavior_tree_rqt.launch 

Choose and modify the prompts

• As the user interface shown, there are one varient input:

  • Task description(key in the requirements you need)

• And four primitive prompts that can direct enable/disable via the select buttons left hand side:(you can see the full context here)

  • Environment descriptions
  • Robot sensing abilities
  • Source behavior trees
  • Existing sub trees

• You can also view the details and edit primitive prompts by using the "view detail and edit" buttons right hand side:

• Click the "Submit" button to start generate behavior tree.

• The detailed demostration can refer this video: watch demo video in youtube