Behavior Tree Generation

Purpose

Turn a natural-language scenario into behavior-tree output using an LLM-driven planner. This is a focused sub-skill of Action and Event Data Generation, packaged as omni.ai.behavior_tree_gen.core (scripted pipeline + API) and omni.ai.behavior_tree_gen.bridge (Kit UI).

Prerequisites

• Installed Isaac Sim with the Action and Event Data Generation app ($ISAAC_SIM_DIR).
• NVIDIA GPU with a current driver (nvidia-smi) for an actual run (offline helper scripts need neither).
• Shell env contract from isaac-sim-orchestrator: $ISAAC_SIM_DIR, $WORKSPACE_DIR.
• $NVIDIA_API_KEY for the chat/embedding models (prepare_runtime fails without it).

Limitations

• Requires a valid NVIDIA API key; prepare_runtime() does not fall back to a local model.
• Strict call order — setup_workspace() → prepare_runtime() (must return success=True) → generate_behavior_tree().
• Bundled example actions (e.g. MoveTo) are transitional: they demonstrate extensibility, not production quality, and can misbehave.
• This generates a behavior tree from text; it is distinct from the hand-authored actor behavior_tree JSON consumed by an Actor SDG (isaacsim.replicator.agent) config.