CLI Runner#
cli_runner.py is the scripted entry point for inspecting
the generation interface, resolving a prompt, building an existing graph spec,
or running the complete pipeline. Run it inside the Isaac Lab-Arena development
container:
python isaaclab_arena_examples/agentic_environment_generation/cli_runner.py \
--mode full \
--prompt "Franka picks up a cube from the maple table and places it into a bowl."
Modes and system stages#
The --mode option selects which parts of the
system overview run:
Mode |
What it does |
System stages |
|---|---|---|
|
Prints the Pydantic JSON schema for |
Inspects the spec contract. |
|
Prints the asset, relation, and task catalogs exposed to the agent and exits. It does not call the model or start Isaac Sim. |
Inspects the environment generation agent inputs. |
|
Prints the background prim tree from |
Inspects background object-reference paths. |
|
Sends the prompt and catalogs to the agent, validates the returned spec,
prints its graph, and writes |
Prompt → agent → spec. |
|
Loads |
Spec → Arena environment → placement and validation → evaluation smoke test. |
|
Resolves a prompt, writes the generated YAML, builds it, and runs a zero-action policy in one process. This is the default. |
Prompt through evaluation, without a manual-edit pause. |
Use resolve followed by build when the generated YAML needs manual
review:
python isaaclab_arena_examples/agentic_environment_generation/cli_runner.py \
--mode resolve \
--prompt "Droid places the mustard bottle in the grey bin."
# Review or edit the generated YAML, then:
python isaaclab_arena_examples/agentic_environment_generation/cli_runner.py \
--mode build \
--env_spec isaaclab_arena_environments/agent_generated/<env_name>.yaml \
--headless
Runner options#
These options are defined by the environment generation runner:
Option |
Purpose |
System stage |
|---|---|---|
|
Selects the phases described above. Default: |
Pipeline selection |
|
Natural-language environment and task description. |
Prompt |
|
Overrides the generation agent’s default model. |
Agent |
|
Selects the configured inference endpoint. See Inference Model and Spec Quality for endpoint defaults and API keys. |
Agent |
|
Sets model sampling temperature. Default: |
Agent |
|
Chooses where generated graph-spec YAML files are written. Default:
|
Spec |
|
Searches SimReady for prompt objects missing from Arena’s asset catalog. |
Agent |
|
Chooses the SimReady search backend. Default: |
Agent |
|
Overrides the S3 root for S3-based SimReady sources. |
Agent |
|
Overrides the hosted USD Search service URL. |
Agent |
|
Limits retained SimReady hits per searched object. Default: |
Agent |
|
Sets the number of zero-action simulation steps. Default: |
Evaluation smoke test |
Build and simulation options#
The runner also inherits Arena and Isaac Lab launcher options. The options most
relevant to build and full are:
Option |
Purpose |
System stage |
|---|---|---|
|
Input graph-spec YAML. Required by |
Spec |
|
Runs Isaac Sim without a viewport. |
Build and evaluation |
|
Sets the number of parallel environments. Default: |
Arena environment |
|
Sets spacing between parallel environments. Default: |
Arena environment |
|
Sets the simulation random seed. Default: |
Build and evaluation |
|
Disables spatial-relation solving. |
Placement solving |
|
Makes object-placement solving reproducible. |
Placement solving |
|
Enables or disables re-placement of pooled objects at reset. |
Placement solving |
|
Uses USD I/O instead of Fabric. |
Arena environment |
|
Builds a mimic environment for demonstration workflows. |
Environment compilation |
|
Selects a physics-backend preset. |
Environment compilation |
Run the script with --help for the complete launcher-specific option list,
including device and livestream settings supplied by Isaac Lab.