Unify Expert Program runtime integration

[yuecideng]

Description

This PR makes EmbodiedEnv the shared execution boundary for declarative Expert Programs and removes duplicate runtime assembly across the semantic-skill and task layers.

Key changes:

• centralize semantic runtime assembly and scene/profile validation for both AtomicSkills and Expert Program execution;
• add registration-owned factories for custom semantic lowerers, including the Open Drawer Slide call;
• add typed planning-failure classification, named trajectory-segment events, and bound skill/call availability queries;
• let EnvSpec inject one immutable Expert Program adapter factory and derive preflight registration metadata from the same owner;
• let EmbodiedEnv.create_demo_segments() accept either static configuration or an episode-level ExpertProgramCfg/CompiledProgram, providing the handoff required by a future MLLM frontend;
• migrate Repeated Pick/Place and Open Drawer to the shared environment path without task-local adapter properties or runtime assembly;
• consolidate completed design material into maintained Sphinx documentation and routed agent context.

HandOver and ActionBank are intentionally unchanged. ActionBank deprecation and GPU-backed physical qualification remain outside this PR.

Dependencies: None.

Refs #471
Refs #474

Type of change

• Enhancement (non-breaking change which improves existing functionality)

Screenshots

Not applicable; this change has no visual UI.

Validation

• black . — 815 files unchanged
• python docs/scripts/check_api_docs.py — 1709/1709 exports documented
• pytest -q tests/sim/atomic_actions -m 'not gpu and not requires_sim' --disable-warnings --maxfail=10 — 564 passed, 4 deselected
• pytest -q tests/sim/skills -m 'not gpu and not requires_sim' --disable-warnings --maxfail=10 — 333 passed, 13 deselected
• pytest -q tests/gym/envs/expert_program tests/gym/envs/test_embodied_env_expert_program.py tests/gym/utils/test_gym_utils.py -m 'not gpu and not requires_sim' --disable-warnings --maxfail=10 — 321 passed, 1 deselected
• pytest -q tests/docs --confcutdir=tests/docs --disable-warnings --maxfail=10 — 20 passed
• git diff --check HEAD^ — passed

GPU and real-simulator trajectory qualification were not run locally.

Checklist

• I have run the black . command to format the code base.
• I have made corresponding changes to the documentation.
• Public API changes are reflected in the API docs (python docs/scripts/check_api_docs.py).
• I have added tests that prove the feature works.
• Dependencies have been updated, if applicable (not applicable; no dependency changes).

[greptile-apps]

Greptile Summary

The PR centralizes Expert Program compilation and runtime assembly behind EmbodiedEnv, with registration-owned adapter and semantic-lowerer factories.

• Adds episode-level Expert Program selection and shared environment lifecycle integration.
• Consolidates semantic runtime assembly, availability diagnostics, planning-failure classification, and trajectory-segment events.
• Migrates the reference Expert Program tasks to registration-owned shared runtime construction.
• Updates tests, API documentation, and maintained architecture guidance while removing superseded design and reporting material.

Confidence Score: 5/5

The PR appears safe to merge because no blocking failure remains in the eligible follow-up review scope.

No blocking failure remains.

Important Files Changed

Filename	Overview
embodichain/lab/gym/envs/embodied_env.py	Adds registration-injected adapter binding, episode-level program selection, and bridge-aware task-success handling.
embodichain/lab/gym/utils/registration.py	Extends environment registration to own and inject an immutable Expert Program adapter factory.
embodichain/lab/gym/envs/expert_program/environment.py	Routes Expert Program construction through the shared semantic runtime assembly and registration-owned lowerers.
embodichain/lab/gym/envs/expert_program/extensions.py	Defines immutable declarations and factories for registered semantic lowerers.
embodichain/lab/gym/envs/expert_program/simulation_environment.py	Implements simulation-backed adapter and lowerer creation from the task registration.
embodichain/lab/sim/skills/_assembly.py	Introduces the common provider-free semantic runtime assembly path.
embodichain/lab/sim/atomic_actions/plans.py	Adds typed planning-failure metadata with a compatibility default for unclassified failed plans.
embodichain/lab/sim/atomic_actions/execution.py	Emits named segment events and handles non-retryable planning failures without charging retry budgets.
embodichain/lab/sim/skills/runtime.py	Exposes bound skill and semantic-call availability while using the shared runtime assembly.
embodichain_tasks/embodichain_tasks/expert_program/open_drawer.py	Moves the task-specific Slide lowerer into a registration-owned factory and shared environment path.

Sequence Diagram

sequenceDiagram
participant Caller
participant EnvSpec
participant Env as EmbodiedEnv
participant Adapter
participant Runtime as SkillRuntime
participant Gym
EnvSpec->>Env: construct with adapter factory
Env->>Adapter: create_adapter(live environment)
Caller->>Env: create_demo_segments(program)
Env->>Adapter: compile and create bridge
Adapter->>Runtime: assemble fresh semantic runtime
loop Program segments
Runtime-->>Env: controller action
Env-->>Gym: DemoSegment action
Gym->>Gym: env.step(action)
end
Env-->>Caller: completion mask

Loading

_{Reviews (4): Last reviewed commit: "wip" | Re-trigger Greptile}

[greptile-apps]

Too many files changed for review (125 files, 100 file limit).

Bypass the limit by tagging @greptile-apps to review.