franka_pick_place_code_env and franka_pick_place_multi_code_env reference an unregistered low-level env

Summary

Two registered configurations, franka_pick_place_code_env and franka_pick_place_multi_code_env, declare low_level="franka_cubes_low_level". No environment by that name is ever registered, so selecting either config raises KeyError at construction. The neighbouring franka_robosuite_pick_place_code_env uses franka_robosuite_cubes_low_level, which does exist — this looks like a rename that missed two call sites.

Where

cap/envs/tasks/__init__.py

• line 38 — franka_pick_place_code_env → low_level="franka_cubes_low_level"
• line 73 — franka_pick_place_multi_code_env → low_level="franka_cubes_low_level"
• line 46 — franka_robosuite_pick_place_code_env → low_level="franka_robosuite_cubes_low_level" (correct)

Reproduce

from aspire.sim.cap.envs.tasks import get_config, get_exec_env
from aspire.sim.cap.envs.base import list_envs

for name in ["franka_pick_place_code_env",
             "franka_pick_place_multi_code_env",
             "franka_robosuite_pick_place_code_env"]:
    cfg = get_config(name)
    print(name, cfg.low_level, cfg.low_level in list_envs())
    if cfg.low_level not in list_envs():
        get_exec_env(name)(cfg)

In .venv-robosuite:

franka_pick_place_code_env           franka_cubes_low_level            False
  -> KeyError: "Environment 'franka_cubes_low_level' not registered"
franka_pick_place_multi_code_env     franka_cubes_low_level            False
  -> KeyError: "Environment 'franka_cubes_low_level' not registered"
franka_robosuite_pick_place_code_env franka_robosuite_cubes_low_level  True

Same result in .venv-libero. grep -rn franka_cubes_low_level cap/ env_configs/ returns only these two config declarations — there is no matching register_env anywhere.

Suggested fix

Point both at franka_robosuite_cubes_low_level, matching the sibling config.

Secondary: the test that should catch this skips instead

tests/test_environments.py::check_environment turns exactly this condition into a skip:

if cfg.low_level not in list_envs():
    pytest.skip(f"Low-level environment '{cfg.low_level}' is not registered")

so the suite reports success while never exercising the broken configs. Measured on a fresh clone with ASPIRE_INTEGRATION_REAL=1 and the perception servers up:

• .venv-libero: 10 skipped — nothing executed
• .venv-robosuite: 3 passed, 7 skipped

Of the 7 that skip in both environments, one (r1pro_b1k_low_level) is a legitimate BEHAVIOR-1K gate, two are the franka_cubes_low_level cases above, and four target configurations whose register_exec_env / register_config calls are commented out at cap/envs/tasks/__init__.py:109-142:

"Configuration 'franka_libero_pick_place_code_env' not registered"
"Configuration 'franka_libero_pick_place_code_env_privileged' not registered"
"Configuration 'franka_libero_open_microwave_code_env' not registered"
"Configuration 'franka_libero_open_microwave_code_env_privileged' not registered"

Those four tests can never run as written. Making an unregistered low_level a failure rather than a skip — and deleting or updating the four tests whose registrations were retired — would let this class of defect surface on its own.

Environment

Fresh clone at f4c8939, Linux x86-64, .venv-libero (Python 3.12) and .venv-robosuite (Python 3.10), both built with uv sync --locked per the README.

Start in cap/envs/tasks/init.py at the two franka_pick_place configurations and compare them with the registered robosuite sibling. Update the broken low-level references, then inspect tests/test_environments.py::check_environment so missing low-level environments are not silently skipped. Run the environment tests in the documented virtual environments and verify the two configurations construct successfully.