From 98df4068b376792b06f109839c7ea96078ddc861 Mon Sep 17 00:00:00 2001 From: TATP-233 Date: Mon, 14 Sep 2026 15:20:33 +0800 Subject: [PATCH] chore: prune redundant tests and remove legacy task-factory seam Remove defensive tests whose guarded old interfaces no longer exist, strict-duplicate coverage, tautological/mock-only tests, tests of third-party private internals, and dead fixtures identified by a full tests/ audit (#1572). - delete 9 test files; surgically prune redundant tests/assertions in 22 files, folding unique assertions into the g1 owner contract suite - remove src/unilab/tasks/compatibility.py and its seam: the registry closeout now enforces that legacy EnvCfg -> NpEnv factories cannot coexist with the canonical Manager-Based runtime - replace the outdated private _RslRlVecEnvWrapper copy with the production uni_rl wrapper - update 6 docs pages (en/zh_CN) that cited the deleted obs-alignment test to reference ObservationManager history buffers instead Issue: #1572 --- .../2-user_guide/4-tasks/2-motion_tracking.md | 5 +- .../1-sim_to_real/2-g1_whole_body.md | 3 +- .../1-sim_to_real/8-latency_budget.md | 2 +- .../2-user_guide/4-tasks/2-motion_tracking.md | 3 +- .../1-sim_to_real/2-g1_whole_body.md | 3 +- .../1-sim_to_real/8-latency_budget.md | 2 +- src/unilab/tasks/compatibility.py | 132 ---------- src/unilab/tasks/migration_matrix.py | 2 +- tests/algos/test_appo_runner.py | 17 -- .../test_offpolicy_double_buffer_runner.py | 43 ---- tests/algos/test_offpolicy_dp_sync.py | 14 - tests/algos/test_rsl_rl_runner.py | 78 +----- tests/base/backend/test_drake_batch_pool.py | 74 ------ tests/base/test_body_state_copy.py | 48 ---- tests/base/test_dr_legacy_removed.py | 36 --- tests/base/test_mjwarp_backend.py | 3 - tests/base/test_mjwarp_cuda_graph.py | 243 ------------------ tests/base/test_mjwarp_host_cache.py | 180 ------------- tests/base/test_np_env.py | 22 -- tests/config/test_config_system.py | 14 +- tests/config/test_locomotion_params.py | 30 --- tests/conftest.py | 35 --- .../a2/test_a2_joystick_contract.py | 14 +- .../locomotion/g1/test_g1_owner_contract.py | 23 +- .../go1/test_manager_based_flat_cfg.py | 8 - .../locomotion/go2/test_manager_based_cfg.py | 9 +- .../go2w/test_go2w_manager_based_flat_cfg.py | 7 - tests/envs/locomotion/test_gait_terms.py | 4 - tests/envs/locomotion/test_go2_footstand.py | 8 - tests/envs/test_env_configs.py | 78 ------ tests/envs/test_motion_interpolation.py | 184 +------------ tests/envs/test_stewart.py | 10 +- .../test_reward_injection_integration.py | 30 --- tests/ipc/test_dp_launcher.py | 5 - .../test_observation_buffers_noise.py | 15 -- tests/scripts/test_obs_alignment_g1_wbt.py | 216 ---------------- tests/tasks/test_legacy_task_compatibility.py | 176 ------------- tests/tasks/test_migration_matrix.py | 44 ---- .../test_production_registry_closeout.py | 53 +--- tests/utils/test_mjspec_sensor_compile.py | 26 -- tests/utils/test_utils_package_policy.py | 12 - 41 files changed, 44 insertions(+), 1867 deletions(-) delete mode 100644 src/unilab/tasks/compatibility.py delete mode 100644 tests/base/test_body_state_copy.py delete mode 100644 tests/base/test_dr_legacy_removed.py delete mode 100644 tests/base/test_mjwarp_cuda_graph.py delete mode 100644 tests/base/test_mjwarp_host_cache.py delete mode 100644 tests/integration/test_reward_injection_integration.py delete mode 100644 tests/scripts/test_obs_alignment_g1_wbt.py delete mode 100644 tests/tasks/test_legacy_task_compatibility.py delete mode 100644 tests/tasks/test_migration_matrix.py delete mode 100644 tests/utils/test_mjspec_sensor_compile.py diff --git a/docs/sphinx/source/en/2-user_guide/4-tasks/2-motion_tracking.md b/docs/sphinx/source/en/2-user_guide/4-tasks/2-motion_tracking.md index 7b402c533..5bc28206c 100644 --- a/docs/sphinx/source/en/2-user_guide/4-tasks/2-motion_tracking.md +++ b/docs/sphinx/source/en/2-user_guide/4-tasks/2-motion_tracking.md @@ -65,8 +65,9 @@ actor keeps the command and anchor-orientation terms at one step while the `base_ang_vel`, `joint_pos`, `joint_vel`, and `actions` terms declare `history_length: 5`. `ObservationManager` owns and flattens those per-term histories; the actor uses the configured encoder-biased joint-position term while -the critic keeps the clean term. Per-term oldest-first ordering is guarded by -`tests/scripts/test_obs_alignment_g1_wbt.py`; the hardware-side contract is +the critic keeps the clean term. Per-term oldest-first ordering is guaranteed by +the `ObservationManager` per-term history buffers +(`tests/managers/test_observation_buffers_noise.py`); the hardware-side contract is documented in the sim-to-real deployment guide. When a Motrix sim2sim replay needs a checkpoint from another log root, pass the absolute path through `uv run eval`: diff --git a/docs/sphinx/source/en/3-deployment/1-sim_to_real/2-g1_whole_body.md b/docs/sphinx/source/en/3-deployment/1-sim_to_real/2-g1_whole_body.md index 60d087cee..fe3f0bf9f 100644 --- a/docs/sphinx/source/en/3-deployment/1-sim_to_real/2-g1_whole_body.md +++ b/docs/sphinx/source/en/3-deployment/1-sim_to_real/2-g1_whole_body.md @@ -97,7 +97,8 @@ within the term, and terms are concatenated in declaration order: The motion command contributes the reference joint position and velocity (`29 + 29`) ahead of the observation terms. Per-term oldest-first ordering is -guarded by `tests/scripts/test_obs_alignment_g1_wbt.py`; mirror that ordering +guaranteed by the `ObservationManager` per-term history buffers +(`tests/managers/test_observation_buffers_noise.py`); mirror that ordering on hardware or the policy reads a permuted vector. ## 3. Actuator interface diff --git a/docs/sphinx/source/en/3-deployment/1-sim_to_real/8-latency_budget.md b/docs/sphinx/source/en/3-deployment/1-sim_to_real/8-latency_budget.md index 6cecd56ce..8a373e25c 100644 --- a/docs/sphinx/source/en/3-deployment/1-sim_to_real/8-latency_budget.md +++ b/docs/sphinx/source/en/3-deployment/1-sim_to_real/8-latency_budget.md @@ -10,7 +10,7 @@ budgets as robot-specific measurements, not UniLab defaults. | --- | --- | --- | | One-step action delay | Manager action term `simulate_action_latency` declarations in task owners | Executes the previous action instead of the current action. | | G1 WBT observation history | Per-term `history_length` in `src/unilab/conf/sac/task/g1_wbt_obs/mujoco.yaml` | Per-term history for `base_ang_vel`, `joint_pos`, `joint_vel`, and `actions`. | -| Obs history ordering guard | `tests/scripts/test_obs_alignment_g1_wbt.py` | Asserts per-term oldest-first flatten for the G1 WBT actor obs. | +| Obs history ordering | `ObservationManager` per-term history buffers (`tests/managers/test_observation_buffers_noise.py`) | Per-term oldest-first flatten for the G1 WBT actor obs. | ## Action Latency diff --git a/docs/sphinx/source/zh_CN/2-user_guide/4-tasks/2-motion_tracking.md b/docs/sphinx/source/zh_CN/2-user_guide/4-tasks/2-motion_tracking.md index c8b279f49..c4acb0154 100644 --- a/docs/sphinx/source/zh_CN/2-user_guide/4-tasks/2-motion_tracking.md +++ b/docs/sphinx/source/zh_CN/2-user_guide/4-tasks/2-motion_tracking.md @@ -61,7 +61,8 @@ uv run train --algo sac --task g1_wbt_obs --sim mujoco training.use_amp=true orientation term 保持单步，`base_ang_vel`、`joint_pos`、`joint_vel` 和 `actions` term 分别声明 `history_length: 5`。这些逐项历史由 `ObservationManager` 维护并展开；actor 使用配置中的 encoder-biased joint-position term，critic 则保留 clean term。逐项最旧 -优先顺序由 `tests/scripts/test_obs_alignment_g1_wbt.py` 守护；硬件侧契约见仿真到真机 +优先顺序由 `ObservationManager` 的逐项历史缓冲实现保证 +（`tests/managers/test_observation_buffers_noise.py`）；硬件侧契约见仿真到真机 部署指南。当 Motrix sim2sim 回放需要引用其他日志根目录下的 checkpoint 时，用 `uv run eval` 透传绝对路径： diff --git a/docs/sphinx/source/zh_CN/3-deployment/1-sim_to_real/2-g1_whole_body.md b/docs/sphinx/source/zh_CN/3-deployment/1-sim_to_real/2-g1_whole_body.md index 52567a1b3..d1ced945e 100644 --- a/docs/sphinx/source/zh_CN/3-deployment/1-sim_to_real/2-g1_whole_body.md +++ b/docs/sphinx/source/zh_CN/3-deployment/1-sim_to_real/2-g1_whole_body.md @@ -91,7 +91,8 @@ Actor 观测宽度是 `env.observations.actor.terms` 下各项 `dim * history_le ``` motion command 在观测项之前贡献参考关节位置与速度（`29 + 29`）。逐项的最旧优先 -顺序由 `tests/scripts/test_obs_alignment_g1_wbt.py` 守护；硬件侧必须镜像该顺序， +顺序由 `ObservationManager` 的逐项历史缓冲实现保证 +（`tests/managers/test_observation_buffers_noise.py`）；硬件侧必须镜像该顺序， 否则策略读到的是被置换过的向量。 ## 3. 执行器接口 diff --git a/docs/sphinx/source/zh_CN/3-deployment/1-sim_to_real/8-latency_budget.md b/docs/sphinx/source/zh_CN/3-deployment/1-sim_to_real/8-latency_budget.md index b2acebb6c..33d8814d9 100644 --- a/docs/sphinx/source/zh_CN/3-deployment/1-sim_to_real/8-latency_budget.md +++ b/docs/sphinx/source/zh_CN/3-deployment/1-sim_to_real/8-latency_budget.md @@ -9,7 +9,7 @@ | --- | --- | --- | | 单步动作延迟 | task owner 中 Manager action term 的 `simulate_action_latency` 声明 | 执行上一步动作而非当前动作。 | | G1 WBT 观测历史 | `src/unilab/conf/sac/task/g1_wbt_obs/mujoco.yaml` 中逐 term 的 `history_length` | 为 `base_ang_vel`、`joint_pos`、`joint_vel` 与 `actions` 提供逐项历史。 | -| 观测历史顺序守护 | `tests/scripts/test_obs_alignment_g1_wbt.py` | 断言 G1 WBT actor 观测按逐项最旧优先展平。 | +| 观测历史顺序 | `ObservationManager` 逐项历史缓冲（`tests/managers/test_observation_buffers_noise.py`） | G1 WBT actor 观测按逐项最旧优先展平。 | ## 动作延迟 diff --git a/src/unilab/tasks/compatibility.py b/src/unilab/tasks/compatibility.py deleted file mode 100644 index a33106b20..000000000 --- a/src/unilab/tasks/compatibility.py +++ /dev/null @@ -1,132 +0,0 @@ -"""Internal, cold-path compatibility seam for legacy task factories. - -This module is intentionally task-owned and is not part of the base registry -contract. It only admits legacy factories that already use UniLab's -``EnvCfg -> NpEnv`` lifecycle; it does not provide a fallback runtime. -""" - -from __future__ import annotations - -from dataclasses import dataclass -from enum import Enum -from typing import Generic, Protocol, TypeVar - -from unilab.base.base import ABEnv, EnvCfg -from unilab.base.np_env import NpEnv - - -class CompatibilityStatus(str, Enum): - """Documented outcome for one legacy compatibility boundary.""" - - ADAPTED = "Adapted" - UNSUPPORTED = "Unsupported" - - -@dataclass(frozen=True) -class LegacyTaskCompatibility: - """Immutable compatibility evidence attached to one task-family seam.""" - - task_family: str - status: CompatibilityStatus - reason: str - - def __post_init__(self) -> None: - if not self.task_family.strip(): - raise ValueError("legacy compatibility task_family must be non-empty") - if not self.reason.strip(): - raise ValueError("legacy compatibility reason must be non-empty") - - -TCfg_contra = TypeVar("TCfg_contra", bound=EnvCfg, contravariant=True) - - -class LegacyEnvFactory(Protocol[TCfg_contra]): - """Existing registry-shaped legacy factory admitted by this seam.""" - - def __call__( - self, - cfg: TCfg_contra, - *, - num_envs: int = 1, - backend_type: str = "mujoco", - ) -> ABEnv: ... - - -@dataclass(frozen=True) -class LegacyFactoryAdapter(Generic[TCfg_contra]): - """Validate one legacy factory at the existing env construction boundary.""" - - factory: LegacyEnvFactory[TCfg_contra] - compatibility: LegacyTaskCompatibility - - def __post_init__(self) -> None: - if self.compatibility.status is not CompatibilityStatus.ADAPTED: - raise ValueError("LegacyFactoryAdapter compatibility status must be Adapted") - - def __call__( - self, - cfg: TCfg_contra, - *, - num_envs: int = 1, - backend_type: str = "mujoco", - ) -> NpEnv: - family = self.compatibility.task_family - if not isinstance(cfg, EnvCfg): - raise TypeError( - f"Legacy task family '{family}' expected EnvCfg, received {type(cfg).__name__}" - ) - - env = self.factory(cfg, num_envs=num_envs, backend_type=backend_type) - if not isinstance(env, ABEnv): - raise TypeError( - f"Legacy task family '{family}' factory returned {type(env).__name__}, " - "expected ABEnv" - ) - if not isinstance(env, NpEnv): - raise TypeError( - f"Legacy task family '{family}' compatibility is Unsupported: " - f"{type(env).__name__} does not use the NpEnv lifecycle" - ) - return env - - -def adapt_legacy_factory( - factory: LegacyEnvFactory[TCfg_contra], - *, - task_family: str, - reason: str, -) -> LegacyFactoryAdapter[TCfg_contra]: - """Mark and wrap an existing ``EnvCfg -> NpEnv`` task factory. - - The wrapper runs only while the registry constructs an environment. It - forwards the registry's fixed arguments exactly once and rejects any - other config or runtime shape instead of probing or falling back. - """ - - if not callable(factory): - raise TypeError(f"legacy task family '{task_family}' factory must be callable") - compatibility = LegacyTaskCompatibility( - task_family=task_family, - status=CompatibilityStatus.ADAPTED, - reason=reason, - ) - return LegacyFactoryAdapter(factory=factory, compatibility=compatibility) - - -def unsupported_legacy_task(*, task_family: str, reason: str) -> LegacyTaskCompatibility: - """Record an explicit unsupported surface without creating a factory.""" - - return LegacyTaskCompatibility( - task_family=task_family, - status=CompatibilityStatus.UNSUPPORTED, - reason=reason, - ) - - -__all__ = [ - "CompatibilityStatus", - "LegacyFactoryAdapter", - "LegacyTaskCompatibility", - "adapt_legacy_factory", - "unsupported_legacy_task", -] diff --git a/src/unilab/tasks/migration_matrix.py b/src/unilab/tasks/migration_matrix.py index 366868f73..0e190ad3c 100644 --- a/src/unilab/tasks/migration_matrix.py +++ b/src/unilab/tasks/migration_matrix.py @@ -11,7 +11,7 @@ from typing import Literal MigrationStatus = Literal["Compatible", "Adapted"] -MigrationTarget = Literal["complete", "mba", "compatibility"] +MigrationTarget = Literal["complete", "mba"] @dataclass(frozen=True) diff --git a/tests/algos/test_appo_runner.py b/tests/algos/test_appo_runner.py index 0ab1ad143..496979fd1 100644 --- a/tests/algos/test_appo_runner.py +++ b/tests/algos/test_appo_runner.py @@ -26,23 +26,6 @@ from unilab.structured_configs import APPOConfig -@pytest.mark.slow -def test_appo_runner_init_no_crash(mock_env_name): - cfg = APPOConfig().to_dict() - cfg["num_envs"] = 4 - cfg["steps_per_env"] = 4 - - runner = APPORunner( - env_name=mock_env_name, - env_factory=registry_env_factory(mock_env_name, "mujoco"), - env_cfg_overrides={}, - rl_cfg=cfg, - num_envs=4, - steps_per_env=4, - ) - runner.close() - - @pytest.mark.slow @pytest.mark.parametrize("env_name", ["Go2JoystickFlat"]) def test_appo_runner_learn_two_iterations(env_name): diff --git a/tests/algos/test_offpolicy_double_buffer_runner.py b/tests/algos/test_offpolicy_double_buffer_runner.py index 733c36ee7..a1dd5706c 100644 --- a/tests/algos/test_offpolicy_double_buffer_runner.py +++ b/tests/algos/test_offpolicy_double_buffer_runner.py @@ -11,7 +11,6 @@ import pytest from hydra import compose, initialize_config_dir from hydra.core.global_hydra import GlobalHydra -from hydra.errors import ConfigCompositionException from uni_rl.ipc.dp_launcher import UNILAB_DP_LOG_DIR, UNILAB_DP_RANK, UNILAB_DP_WORLD_SIZE _ROOT = Path(__file__).parent.parent.parent @@ -83,35 +82,6 @@ def test_offpolicy_config_has_one_replay_path(): cfg = _offpolicy_cfg() assert cfg.training.replay_prefetch_mode == "one_tick" assert cfg.training.env_steps_per_sync == 1 - assert "env_steps_per_sync" not in cfg.algo - assert "inference_owner" not in cfg.training - assert "collector_infer_device" not in cfg.training - assert "no_sync_collection" not in cfg.training - assert "replay_pipeline" not in cfg.training - assert "verbose_metrics" not in cfg.training - assert "replay_pack_layout" not in cfg.training - assert "replay_pack_executor" not in cfg.training - assert "replay_h2d_submitter" not in cfg.training - - -@pytest.mark.parametrize( - "override", - [ - "training.replay_pipeline=cpu_pinned_double_buffer", - "training.verbose_metrics=true", - "training.num_gpus=2", - "training.multi_gpu_sync_mode=sync_sgd", - "training.multi_gpu_sync_interval=2", - "training.device=cuda", - "training.inference_owner=collector", - "training.collector_infer_device=cpu", - "training.no_sync_collection=true", - "algo.env_steps_per_sync=2", - ], -) -def test_removed_offpolicy_options_fail_hydra_compose(override: str): - with pytest.raises(ConfigCompositionException, match="Could not override"): - _offpolicy_cfg([override]) @pytest.mark.parametrize("mode", ["invalid_mode", "same_tick"]) @@ -160,11 +130,6 @@ def test_sac_dispatch_constructs_unique_runner(monkeypatch: pytest.MonkeyPatch): assert runner.kwargs["algo_type"] == "sac" assert runner.kwargs["device"] == "cuda:0" assert runner.kwargs["replay_prefetch_mode"] == "one_tick" - assert "inference_owner" not in runner.kwargs - assert "collector_infer_device" not in runner.kwargs - assert "sync_collection" not in runner.kwargs - assert "replay_pipeline" not in runner.kwargs - assert "verbose_metrics" not in runner.kwargs assert runner.kwargs["learner"].kwargs == { "device": "cuda:0", "obs_dim": 4, @@ -252,10 +217,6 @@ def test_td3_dispatch_constructs_unique_runner(monkeypatch: pytest.MonkeyPatch): assert isinstance(runner, _FakeRunner) assert runner.kwargs["algo_type"] == "td3" assert runner.kwargs["device"] == "cuda:0" - assert "inference_owner" not in runner.kwargs - assert "collector_infer_device" not in runner.kwargs - assert "sync_collection" not in runner.kwargs - assert "replay_pipeline" not in runner.kwargs assert runner.kwargs["replay_prefetch_mode"] == "one_tick" nan_guard_cfg = runner.kwargs["nan_guard_cfg"] assert nan_guard_cfg.enabled is True @@ -326,10 +287,6 @@ def test_flashsac_dispatch_constructs_unique_runner(monkeypatch: pytest.MonkeyPa assert runner.kwargs["algo_type"] == "flashsac" assert runner.kwargs["device"] == "cuda:0" assert runner.kwargs["replay_prefetch_mode"] == "one_tick" - assert "inference_owner" not in runner.kwargs - assert "collector_infer_device" not in runner.kwargs - assert "sync_collection" not in runner.kwargs - assert "replay_pipeline" not in runner.kwargs def test_flashsac_n_step_is_rejected(): diff --git a/tests/algos/test_offpolicy_dp_sync.py b/tests/algos/test_offpolicy_dp_sync.py index 3dd427dac..b4ca8ef37 100644 --- a/tests/algos/test_offpolicy_dp_sync.py +++ b/tests/algos/test_offpolicy_dp_sync.py @@ -153,15 +153,6 @@ def test_learner_without_initial_sync_tensors_fails_with_type_error(): runner._dp_init_broadcast() -def test_close_closes_dp_sync_idempotently(): - dp_sync = _FakeDpSync() - runner = _runner_with(_SyncLearner(), dp_sync) - # Avoid the full AsyncRunner.close(); only the dp_sync branch is under test. - runner.dp_sync.close() - runner.dp_sync.close() - assert [name for name, _ in dp_sync.calls] == ["close", "close"] - - def test_close_restores_terminal_and_ipc_before_destroying_process_group(monkeypatch): from uni_rl.offpolicy.runner import OffPolicyRunner @@ -499,11 +490,6 @@ def __init__(self, *args, **kwargs): return runner.kwargs -def test_offpolicy_config_has_no_periodic_parameter_sync_interval(): - cfg = _offpolicy_cfg() - assert "dp_sync_interval" not in cfg.training - - def test_build_runner_single_rank_keeps_dp_sync_none(monkeypatch: pytest.MonkeyPatch): monkeypatch.delenv(UNILAB_DP_RANK, raising=False) kwargs = _build_sac_runner_with_dp_fakes(monkeypatch, []) diff --git a/tests/algos/test_rsl_rl_runner.py b/tests/algos/test_rsl_rl_runner.py index a86020b35..95549844d 100644 --- a/tests/algos/test_rsl_rl_runner.py +++ b/tests/algos/test_rsl_rl_runner.py @@ -21,90 +21,16 @@ ) rsl_rl = pytest.importorskip("rsl_rl") -import numpy as np -import torch -from tensordict import TensorDict -from uni_rl.algos.rsl_rl import normalize_ppo_train_cfg +from uni_rl.algos.rsl_rl import RslRlVecEnvWrapper, normalize_ppo_train_cfg from unilab.base import registry from unilab.base.config_adapter import BackendAdapter from unilab.base.registry import ensure_registries from unilab.structured_configs import PPOConfig -from unilab.utils.tensor import to_torch ensure_registries() -# --------------------------------------------------------------------------- -# Minimal wrapper (same as src/unilab/scripts/train_rsl_rl.py) -# --------------------------------------------------------------------------- - - -class _RslRlVecEnvWrapper: - """Lightweight RSL-RL wrapper for testing.""" - - def __init__(self, env, device="cpu"): - self.env = env - self.cfg = env.cfg - self.device = device - self.num_envs = env.num_envs - self.observation_space = env.observation_space - self.action_space = env.action_space - self.num_obs = int(env.obs_groups_spec["obs"]) - self.num_privileged_obs = int(env.obs_groups_spec.get("critic", self.num_obs)) - self.num_actions = env.action_space.shape[0] - - self.episode_returns = torch.zeros(self.num_envs, device=device) - self.episode_lengths = torch.zeros(self.num_envs, device=device) - self.episode_length_buf = self.episode_lengths - self.max_episode_length = np.ceil(env.cfg.max_episode_seconds / env.cfg.ctrl_dt) - self.reset() - - def _obs_to_tensordict(self, obs: dict[str, np.ndarray]) -> TensorDict: - actor = to_torch(obs["obs"], self.device) - td = {"actor": actor, "policy": actor} - if "critic" in obs: - td["critic"] = to_torch(obs["critic"], self.device) - return TensorDict(td, batch_size=self.num_envs, device=self.device) - - def step(self, actions): - actions_np = ( - actions.detach().cpu().numpy() if isinstance(actions, torch.Tensor) else actions - ) - state = self.env.step(actions_np) - rewards = to_torch(state.reward, self.device) - dones = to_torch(state.terminated | state.truncated, self.device).bool() - self.episode_returns += rewards - self.episode_lengths += 1 - infos = {} - done_idx = torch.nonzero(dones).flatten() - if len(done_idx) > 0: - infos["time_outs"] = to_torch(state.truncated, self.device).bool() - self.episode_returns[done_idx] = 0 - self.episode_lengths[done_idx] = 0 - if "log" in state.info: - infos["log"] = state.info["log"] - return self._obs_to_tensordict(state.obs), rewards, dones, infos - - def reset(self): - if self.env.state is None: - self.env.init_state() - env_indices = np.arange(self.num_envs, dtype=np.int32) - obs_out, _ = self.env.reset(env_indices) - self.episode_returns[:] = 0 - self.episode_lengths[:] = 0 - return self._obs_to_tensordict(obs_out), {} - - def get_observations(self): - assert self.env.state is not None - return self._obs_to_tensordict(self.env.state.obs) - - def get_privileged_observations(self): - assert self.env.state is not None - obs = self.env.state.obs - return to_torch(obs.get("critic", obs["obs"]), self.device) - - # --------------------------------------------------------------------------- # Tests # --------------------------------------------------------------------------- @@ -147,7 +73,7 @@ def test_rsl_rl_ppo_one_iteration( sim_backend="mujoco", env_cfg_override=env_cfg_override, ) - wrapped = _RslRlVecEnvWrapper(env, device="cpu") + wrapped = RslRlVecEnvWrapper(env, device="cpu") cfg = PPOConfig() train_cfg = cfg.to_dict() diff --git a/tests/base/backend/test_drake_batch_pool.py b/tests/base/backend/test_drake_batch_pool.py index 1b2766ec4..59f07bc90 100644 --- a/tests/base/backend/test_drake_batch_pool.py +++ b/tests/base/backend/test_drake_batch_pool.py @@ -109,43 +109,6 @@ def make_go1_pool(nbatch, nthread): """ -def test_batch_import_diagnostic_is_preserved() -> None: - output = _run_clean_python( - """ - import json - - try: - from drake_uni.batch_env import batch_available, batch_import_error - except ImportError as exc: - captured_error = exc - - def batch_available(): - return False - - def batch_import_error(): - return captured_error - - error = batch_import_error() - summary = { - "available": bool(batch_available()), - "error_type": None if error is None else type(error).__name__, - "missing_module": getattr(error, "name", None), - } - print(json.dumps(summary, sort_keys=True)) - """ - ) - summary = json.loads(output.strip().splitlines()[-1]) - if summary["available"]: - assert summary["error_type"] is None - assert summary["missing_module"] is None - else: - assert summary["error_type"] == "ModuleNotFoundError" - assert summary["missing_module"] in { - "drake_uni", - "drake_uni.compiled._drake_env_pool", - } - - def test_batch_backend_mode_rejects_existing_pydrake_module() -> None: output = _run_clean_python( """ @@ -284,41 +247,6 @@ def test_drake_uni_runtime_import_is_lazy_and_pydrake_free() -> None: } -def test_unilab_drake_public_surface_excludes_batch_backend_symbol() -> None: - output = _run_clean_python( - """ - import json - - import unilab.base.backend_factory as backend_root - import unisim.backend.drake as drake_pkg - from unisim.backend.drake import backend as backend_module - - try: - from unisim.backend.drake.backend import DrakeUniBatchBackend # noqa: F401 - except ImportError: - direct_import = "failed" - else: - direct_import = "succeeded" - - summary = { - "direct_import": direct_import, - "root_has_batch": hasattr(backend_root, "DrakeUniBatchBackend"), - "subpackage_has_batch": hasattr(drake_pkg, "DrakeUniBatchBackend"), - "module_has_batch": hasattr(backend_module, "DrakeUniBatchBackend"), - "module_all_has_batch": "DrakeUniBatchBackend" in backend_module.__all__, - } - print(json.dumps(summary, sort_keys=True)) - """ - ) - assert json.loads(output.strip().splitlines()[-1]) == { - "direct_import": "failed", - "module_all_has_batch": False, - "module_has_batch": False, - "root_has_batch": False, - "subpackage_has_batch": False, - } - - @pytest.mark.skipif( not _batch_extension_built(), reason="optional Drake batch extension has not been built", @@ -336,7 +264,6 @@ def test_drake_batch_pool_go1_smoke_shapes_and_time() -> None: output = pool.step(state, 2, control, None, True) sensor_data = output["sensor_data"] summary = { - "forward_removed": not hasattr(pool, "forward"), "state_only_has_sensor_data": "sensor_data" in state_only, "state_shape": list(output["state"].shape), "sensor_shape": list(sensor_data.shape), @@ -355,7 +282,6 @@ def test_drake_batch_pool_go1_smoke_shapes_and_time() -> None: summary = json.loads(output.strip().splitlines()[-1]) assert summary.pop("num_filtered_geometries") > 0 assert summary == { - "forward_removed": True, "has_sensor_data": True, "nthread": 1, "sensor_finite": True, diff --git a/tests/base/test_body_state_copy.py b/tests/base/test_body_state_copy.py deleted file mode 100644 index 28a06e290..000000000 --- a/tests/base/test_body_state_copy.py +++ /dev/null @@ -1,48 +0,0 @@ -"""Focused parity tests for the shared body-state copy kernel.""" - -from __future__ import annotations - -import numpy as np -from unisim.backend.body_state import copy_selected_body_state - - -def _outputs(num_envs: int, num_selected: int) -> tuple[np.ndarray, ...]: - shape = (num_envs, num_selected, 3) - return ( - np.empty(shape, dtype=np.float32), - np.empty((num_envs, num_selected, 4), dtype=np.float32), - np.empty(shape, dtype=np.float32), - np.empty(shape, dtype=np.float32), - ) - - -def test_shared_body_state_copy_kernel_preserves_selection_and_outputs() -> None: - rng = np.random.default_rng(7) - num_envs, num_bodies = 257, 9 - pos = rng.standard_normal((num_envs, num_bodies, 3), dtype=np.float32) - quat = rng.standard_normal((num_envs, num_bodies, 4), dtype=np.float32) - lin_vel = rng.standard_normal((num_envs, num_bodies, 3), dtype=np.float32) - ang_vel = rng.standard_normal((num_envs, num_bodies, 3), dtype=np.float32) - selected = np.asarray([7, 1, 5], dtype=np.intp) - out_pos, out_quat, out_lin_vel, out_ang_vel = _outputs(num_envs, len(selected)) - - copy_selected_body_state( - pos, - quat, - lin_vel, - ang_vel, - selected, - out_pos, - out_quat, - out_lin_vel, - out_ang_vel, - ) - - np.testing.assert_array_equal(out_pos, pos[:, selected]) - np.testing.assert_array_equal(out_quat, quat[:, selected]) - np.testing.assert_array_equal(out_lin_vel, lin_vel[:, selected]) - np.testing.assert_array_equal(out_ang_vel, ang_vel[:, selected]) - # The shared helper intentionally stays NumPy-only so the core package has - # no mandatory Numba dependency. Backend-specific compiled kernels belong - # to their optional adapter extras. - assert not hasattr(copy_selected_body_state, "targetoptions") diff --git a/tests/base/test_dr_legacy_removed.py b/tests/base/test_dr_legacy_removed.py deleted file mode 100644 index dbc990d02..000000000 --- a/tests/base/test_dr_legacy_removed.py +++ /dev/null @@ -1,36 +0,0 @@ -from __future__ import annotations - -import importlib -import subprocess -import sys -from pathlib import Path - -import pytest - - -def test_legacy_dr_protocol_is_removed() -> None: - import unilab.base.np_env as np_env - - assert not hasattr(np_env.NpEnv, "_init_domain_randomization") - with pytest.raises(ModuleNotFoundError, match="No module named 'unilab.dr'"): - importlib.import_module("unilab.dr") - - source = Path(np_env.__file__).read_text(encoding="utf-8") - assert "_dr_manager" not in source - assert "unilab.dr" not in source - - -def test_fresh_import_graph_does_not_load_legacy_dr_manager() -> None: - code = "\n".join( - [ - "import sys", - "import unilab", - "assert 'unilab.dr' not in sys.modules", - "assert 'unilab.dr.manager' not in sys.modules", - "assert 'unilab.dr.provider' not in sys.modules", - ] - ) - result = subprocess.run( - [sys.executable, "-c", code], capture_output=True, text=True, check=False - ) - assert result.returncode == 0, result.stderr diff --git a/tests/base/test_mjwarp_backend.py b/tests/base/test_mjwarp_backend.py index e31378445..0aa94cb21 100644 --- a/tests/base/test_mjwarp_backend.py +++ b/tests/base/test_mjwarp_backend.py @@ -359,9 +359,6 @@ def test_set_state_returns_schema_conformant_timing() -> None: assert timing["set_state_host_cache_refresh_ms"] > 0.0 assert timing["set_state_mask_ms"] == 0.0 assert timing["set_state_pool_reset_ms"] == 0.0 - # Legacy collapsed keys are gone. - assert "set_state_reset_ms" not in timing - assert "set_state_cache_refresh_ms" not in timing empty = backend.set_state( np.asarray([], dtype=np.int32), diff --git a/tests/base/test_mjwarp_cuda_graph.py b/tests/base/test_mjwarp_cuda_graph.py deleted file mode 100644 index ad5ed0e1d..000000000 --- a/tests/base/test_mjwarp_cuda_graph.py +++ /dev/null @@ -1,243 +0,0 @@ -"""Cold-path CUDA graph eligibility, capture, replay, and fallback tests.""" - -from __future__ import annotations - -from typing import Any - -import pytest -from unisim.backend.mjwarp.backend import ( - MjwarpBackend, - _cuda_graph_eligibility, - _reset_scratch_capacity_for_batch, -) - - -class _FakeDevice: - def __init__(self, *, is_cuda: bool = True) -> None: - self.is_cuda = is_cuda - - -class _FakeContext: - def __enter__(self) -> "_FakeContext": - return self - - def __exit__(self, *_args: Any) -> None: - return None - - -class _FakeCapture(_FakeContext): - def __init__(self, graph: str, *, failure: Exception | None = None) -> None: - self.graph = graph - self._failure = failure - - def __enter__(self) -> "_FakeCapture": - if self._failure is not None: - raise self._failure - return self - - -class _FakeWarp: - def __init__( - self, - *, - driver: tuple[int, int] | None = (13, 0), - mempool: bool = True, - fail_capture_index: int | None = None, - ) -> None: - self.driver = driver - self.mempool = mempool - self.fail_capture_index = fail_capture_index - self.capture_count = 0 - self.launches: list[str] = [] - - def get_cuda_driver_version(self) -> tuple[int, int] | None: - return self.driver - - def is_mempool_enabled(self, _device: Any) -> bool: - return self.mempool - - def ScopedDevice(self, _device: Any) -> _FakeContext: # noqa: N802 - return _FakeContext() - - def ScopedCapture(self) -> _FakeCapture: # noqa: N802 - capture_index = self.capture_count - self.capture_count += 1 - failure = ( - RuntimeError("synthetic capture failure") - if capture_index == self.fail_capture_index - else None - ) - return _FakeCapture(f"graph-{capture_index}", failure=failure) - - def capture_launch(self, graph: str) -> None: - self.launches.append(graph) - - -class _FakeMujocoWarp: - def __init__(self) -> None: - self.step_calls = 0 - self.forward_calls = 0 - self.reset_calls = 0 - - def step(self, _model: Any, _data: Any) -> None: - self.step_calls += 1 - - def forward(self, _model: Any, _data: Any) -> None: - self.forward_calls += 1 - - def reset_data(self, _model: Any, _data: Any, *, reset: Any) -> None: - del reset - self.reset_calls += 1 - - -def _backend(warp: _FakeWarp, mujoco_warp: _FakeMujocoWarp) -> MjwarpBackend: - backend = MjwarpBackend.__new__(MjwarpBackend) - backend._warp = warp - backend._mujoco_warp = mujoco_warp - backend._device_model = object() - backend._device_data = object() - backend._reset_mask_device = object() - backend._reset_scratch_capacity = 0 - backend._reset_scratch_data = None - return backend - - -@pytest.mark.parametrize( - ("device", "driver", "mempool", "expected_reason"), - [ - (_FakeDevice(is_cuda=False), (13, 0), True, "not CUDA"), - (_FakeDevice(), None, True, "unavailable"), - (_FakeDevice(), (12, 3), True, "older than 12.4"), - (_FakeDevice(), (12, 4), False, "mempool is disabled"), - ], -) -def test_cuda_graph_eligibility_fails_closed( - device: _FakeDevice, - driver: tuple[int, int] | None, - mempool: bool, - expected_reason: str, -) -> None: - eligible, reason = _cuda_graph_eligibility( - _FakeWarp(driver=driver, mempool=mempool), - device, - ) - - assert eligible is False - assert reason is not None and expected_reason in reason - - -def test_cuda_graph_capture_replays_fixed_address_operations() -> None: - warp = _FakeWarp(driver=(12, 4), mempool=True) - mujoco_warp = _FakeMujocoWarp() - backend = _backend(warp, mujoco_warp) - - backend._initialize_cuda_graphs(_FakeDevice()) - - assert backend._cuda_graph_enabled is True - assert backend._cuda_graph_disable_reason is None - assert (backend._step_graph, backend._forward_graph, backend._reset_graph) == ( - "graph-0", - "graph-1", - "graph-2", - ) - capture_calls = ( - mujoco_warp.step_calls, - mujoco_warp.forward_calls, - mujoco_warp.reset_calls, - ) - - backend._execute_device_steps(3) - backend._execute_device_reset() - backend._execute_device_forward() - - assert warp.launches == ["graph-0", "graph-0", "graph-0", "graph-2", "graph-1"] - assert ( - mujoco_warp.step_calls, - mujoco_warp.forward_calls, - mujoco_warp.reset_calls, - ) == capture_calls - - -def test_cuda_graph_capture_includes_materialized_reset_scratch() -> None: - warp = _FakeWarp(driver=(12, 4), mempool=True) - mujoco_warp = _FakeMujocoWarp() - backend = _backend(warp, mujoco_warp) - backend._reset_scratch_capacity = 4 - backend._reset_scratch_data = object() - backend._reset_scratch_mask_device = object() - - backend._initialize_cuda_graphs(_FakeDevice()) - - assert backend._cuda_graph_enabled is True - assert backend._reset_scratch_reset_graph == "graph-3" - assert backend._reset_scratch_forward_graph == "graph-4" - assert mujoco_warp.reset_calls == 2 - assert mujoco_warp.forward_calls == 2 - assert backend._can_use_reset_scratch(4) is True - assert backend._can_use_reset_scratch(5) is False - - -@pytest.mark.parametrize( - ("num_envs", "expected_capacity"), - [ - (512, 0), - (1024, 128), - (2048, 128), - (4096, 256), - (8192, 512), - (16384, 512), - ], -) -def test_reset_scratch_capacity_scales_with_batch_and_stays_bounded( - num_envs: int, expected_capacity: int -) -> None: - assert _reset_scratch_capacity_for_batch(num_envs) == expected_capacity - - -def test_ineligible_cuda_graph_warns_and_uses_eager_operations() -> None: - warp = _FakeWarp(driver=(12, 3), mempool=True) - mujoco_warp = _FakeMujocoWarp() - backend = _backend(warp, mujoco_warp) - - with pytest.warns(RuntimeWarning, match="older than 12.4"): - backend._initialize_cuda_graphs(_FakeDevice()) - - backend._execute_device_steps(2) - backend._execute_device_reset() - backend._execute_device_forward() - - assert backend._cuda_graph_enabled is False - assert warp.capture_count == 0 - assert warp.launches == [] - assert mujoco_warp.step_calls == 2 - assert mujoco_warp.reset_calls == 1 - assert mujoco_warp.forward_calls == 1 - - -def test_cuda_graph_capture_failure_atomically_falls_back_to_eager() -> None: - warp = _FakeWarp(fail_capture_index=1) - mujoco_warp = _FakeMujocoWarp() - backend = _backend(warp, mujoco_warp) - - with pytest.warns(RuntimeWarning, match="synthetic capture failure"): - backend._initialize_cuda_graphs(_FakeDevice()) - - assert backend._cuda_graph_enabled is False - assert backend._step_graph is None - assert backend._forward_graph is None - assert backend._reset_graph is None - assert "synthetic capture failure" in backend._cuda_graph_disable_reason - captured_calls = ( - mujoco_warp.step_calls, - mujoco_warp.forward_calls, - mujoco_warp.reset_calls, - ) - - backend._execute_device_steps(2) - backend._execute_device_reset() - backend._execute_device_forward() - - assert warp.launches == [] - assert mujoco_warp.step_calls == captured_calls[0] + 2 - assert mujoco_warp.forward_calls == captured_calls[1] + 1 - assert mujoco_warp.reset_calls == captured_calls[2] + 1 diff --git a/tests/base/test_mjwarp_host_cache.py b/tests/base/test_mjwarp_host_cache.py deleted file mode 100644 index 9a8557188..000000000 --- a/tests/base/test_mjwarp_host_cache.py +++ /dev/null @@ -1,180 +0,0 @@ -"""Owner-level tests for MJWarp's explicit pinned host-cache barrier.""" - -from __future__ import annotations - -from types import SimpleNamespace -from typing import Any - -import numpy as np -from unisim.backend.mjwarp.backend import MjwarpBackend - - -class _FakeStorage: - def __init__(self, array: np.ndarray) -> None: - self.array = array - - def numpy(self) -> np.ndarray: - return self.array - - -class _FakeWarp: - def __init__(self) -> None: - self.empty_calls: list[dict[str, Any]] = [] - self.events: list[tuple[Any, ...]] = [] - - def empty(self, shape: tuple[int, ...], **kwargs: Any) -> _FakeStorage: - self.empty_calls.append({"shape": shape, **kwargs}) - return _FakeStorage(np.empty(shape, dtype=np.float32)) - - def copy(self, destination: Any, source: Any) -> None: - self.events.append(("copy", destination, source)) - - -def _backend(warp: _FakeWarp) -> MjwarpBackend: - backend = object.__new__(MjwarpBackend) - backend._warp = warp - return backend - - -def test_allocate_host_cache_uses_pinned_cpu_storage() -> None: - warp = _FakeWarp() - backend = _backend(warp) - device_array = SimpleNamespace(shape=(4, 7), dtype="float32") - - storage, cache = backend._allocate_pinned_host_cache(device_array) - - assert warp.empty_calls == [ - {"shape": (4, 7), "dtype": "float32", "device": "cpu", "pinned": True} - ] - assert cache is storage.array - - -def test_refresh_host_cache_batches_all_downloads_before_sync() -> None: - warp = _FakeWarp() - backend = _backend(warp) - backend._device_data = SimpleNamespace( - qpos="device-qpos", qvel="device-qvel", sensordata="device-sensor" - ) - backend._qpos_cache_storage = "host-qpos" - backend._qvel_cache_storage = "host-qvel" - backend._sensor_cache_storage = "host-sensor" - backend._synchronize = lambda: warp.events.append(("sync",)) # type: ignore[method-assign] - - backend._refresh_host_cache() - - assert warp.events == [ - ("copy", "host-qpos", "device-qpos"), - ("copy", "host-qvel", "device-qvel"), - ("copy", "host-sensor", "device-sensor"), - ("sync",), - ] - - -def test_refresh_reset_scratch_cache_only_scatters_selected_rows() -> None: - warp = _FakeWarp() - backend = _backend(warp) - scratch_values = np.arange(12, dtype=np.float32).reshape(4, 3) - scratch_storage = _FakeStorage(np.empty_like(scratch_values)) - backend._reset_scratch_data = SimpleNamespace(sensordata=scratch_values) - backend._reset_scratch_sensor_storage = scratch_storage - backend._reset_scratch_sensor_cache = scratch_storage.array - backend._sensor_cache = np.full((6, 3), -1.0, dtype=np.float32) - backend._download = lambda source, destination: np.copyto( # type: ignore[method-assign] - destination.array, source - ) - backend._synchronize = lambda: warp.events.append(("sync",)) # type: ignore[method-assign] - - backend._refresh_reset_scratch_cache(np.asarray([4, 1], dtype=np.int32)) - - np.testing.assert_array_equal(backend._sensor_cache[4], scratch_values[0]) - np.testing.assert_array_equal(backend._sensor_cache[1], scratch_values[1]) - np.testing.assert_array_equal( - backend._sensor_cache[[0, 2, 3, 5]], - np.full((4, 3), -1.0, dtype=np.float32), - ) - assert warp.events == [("sync",)] - - -def test_host_reset_routes_small_row_sets_through_scratch() -> None: - warp = _FakeWarp() - backend = _backend(warp) - backend._reset_mask_host = np.zeros(8, dtype=np.bool_) - backend._reset_mask_device = "main-mask" - backend._device_data = SimpleNamespace(qpos="main-qpos", qvel="main-qvel") - backend._time_cache = np.ones(8, dtype=np.float32) - events: list[tuple[Any, ...]] = [] - backend._can_use_reset_scratch = lambda _count: True # type: ignore[method-assign] - backend._upload = lambda target, source: events.append( # type: ignore[method-assign] - ("upload", target, np.asarray(source).copy()) - ) - backend._execute_device_reset = lambda: events.append( # type: ignore[method-assign] - ("main-reset",) - ) - backend._execute_reset_scratch_forward = ( # type: ignore[method-assign] - lambda qpos, qvel: events.append(("scratch-forward", qpos.copy(), qvel.copy())) - ) - backend._execute_device_forward = lambda: events.append( # type: ignore[method-assign] - ("main-forward",) - ) - backend._synchronize = lambda: events.append(("sync",)) # type: ignore[method-assign] - backend._refresh_reset_scratch_cache = ( # type: ignore[method-assign] - lambda rows: events.append(("scratch-refresh", rows.copy())) - ) - backend._refresh_host_cache = lambda: events.append( # type: ignore[method-assign] - ("main-refresh",) - ) - rows = np.asarray([6, 2], dtype=np.int32) - full_qpos = np.zeros((8, 3), dtype=np.float32) - full_qvel = np.zeros((8, 2), dtype=np.float32) - reset_qpos = np.ones((2, 3), dtype=np.float32) - reset_qvel = np.ones((2, 2), dtype=np.float32) - - backend._execute_host_reset(rows, full_qpos, full_qvel, reset_qpos, reset_qvel) - - assert np.flatnonzero(backend._reset_mask_host).tolist() == [2, 6] - assert backend._time_cache[rows].tolist() == [0.0, 0.0] - assert [event[0] for event in events] == [ - "upload", - "main-reset", - "upload", - "upload", - "scratch-forward", - "sync", - "scratch-refresh", - ] - assert not any(event[0] in {"main-forward", "main-refresh"} for event in events) - - -def test_row_body_getters_gather_selected_rows_without_full_batch_reads() -> None: - """MJWarp's partial-reset getters must not materialize the full env batch.""" - backend = object.__new__(MjwarpBackend) - backend._body_id_to_tracked_idx = np.asarray([2, 0, 1], dtype=np.intp) - backend._tracked_pos_w_all = np.arange(4 * 3 * 3, dtype=np.float32).reshape(4, 3, 3) - backend._tracked_quat_w_all = np.arange(4 * 3 * 4, dtype=np.float32).reshape(4, 3, 4) - backend._tracked_linvel_w_all = np.arange(4 * 3 * 3, dtype=np.float32).reshape(4, 3, 3) + 1000 - backend._tracked_angvel_w_all = np.arange(4 * 3 * 3, dtype=np.float32).reshape(4, 3, 3) + 2000 - - def fail_full_getter(*_args: Any, **_kwargs: Any) -> Any: - raise AssertionError("row getter must not call a full-batch getter") - - backend.get_body_pos_w = fail_full_getter # type: ignore[method-assign] - backend.get_body_quat_w = fail_full_getter # type: ignore[method-assign] - backend.get_body_lin_vel_w = fail_full_getter # type: ignore[method-assign] - backend.get_body_ang_vel_w = fail_full_getter # type: ignore[method-assign] - - rows = np.asarray([3, 1, 3], dtype=np.int32) - body_ids = np.asarray([1, 2], dtype=np.int32) - mapped = np.asarray([0, 1], dtype=np.intp) - expected_index = (rows[:, None], mapped) - - pose_pos, pose_quat = backend.get_body_pose_w_rows(rows, body_ids) - np.testing.assert_array_equal(pose_pos, backend._tracked_pos_w_all[expected_index]) - np.testing.assert_array_equal(pose_quat, backend._tracked_quat_w_all[expected_index]) - np.testing.assert_array_equal( - backend.get_body_lin_vel_w_rows(rows, body_ids), - backend._tracked_linvel_w_all[expected_index], - ) - np.testing.assert_array_equal( - backend.get_body_ang_vel_w_rows(rows, body_ids), - backend._tracked_angvel_w_all[expected_index], - ) diff --git a/tests/base/test_np_env.py b/tests/base/test_np_env.py index 505f070e2..f6c6937a7 100644 --- a/tests/base/test_np_env.py +++ b/tests/base/test_np_env.py @@ -229,28 +229,6 @@ def test_obs_is_dict(self): assert "obs" in state.obs assert "critic" in state.obs - def test_terminated_and_truncated_are_separate_signals(self): - state = NpEnvState( - obs={"a": np.zeros((3, 1))}, - reward=np.zeros(3), - terminated=np.array([True, False, False]), - truncated=np.array([False, False, True]), - info={}, - ) - np.testing.assert_array_equal(state.terminated, [True, False, False]) - np.testing.assert_array_equal(state.truncated, [False, False, True]) - np.testing.assert_array_equal(state.terminated | state.truncated, [True, False, True]) - - def test_terminated_and_truncated_can_both_be_true(self): - state = NpEnvState( - obs={"a": np.zeros((1, 1))}, - reward=np.zeros(1), - terminated=np.array([True]), - truncated=np.array([True]), - info={}, - ) - assert (state.terminated | state.truncated)[0] is np.True_ - def test_replace_preserves_type(self): obs = {"obs": np.zeros((2, 3))} state = NpEnvState( diff --git a/tests/config/test_config_system.py b/tests/config/test_config_system.py index 4d7a47cbf..5be759f93 100644 --- a/tests/config/test_config_system.py +++ b/tests/config/test_config_system.py @@ -132,14 +132,11 @@ def test_algo_config_composes(algo_dir: str, config_name: str): assert cfg.training.sim_backend == "mujoco" -def test_task_files_keep_full_identity_without_hidden_backend_marker(): +def test_backend_task_files_keep_full_identity(): for path in sorted(CONF_DIR.glob("*/task/**/*.yaml")): cfg = OmegaConf.load(path) cfg_dict_raw = OmegaConf.to_container(cfg, resolve=True) or {} assert isinstance(cfg_dict_raw, dict) - assert "_selected_sim_backend" not in cfg_dict_raw, ( - f"task has hidden backend marker: {path}" - ) if path.stem not in _BACKENDS: continue training_raw = cfg_dict_raw.get("training", {}) @@ -167,15 +164,6 @@ def test_supported_task_composes( _assert_reward_populated(cfg, task_file) -def test_offpolicy_g1_walk_flat_motrix_sac_preserves_backend_overrides(): - cfg = _compose("sac", overrides=["task=g1_walk_flat/motrix"]) - - assert cfg.algo.num_envs == 2048 - assert cfg.algo.max_iterations == 5000 - assert cfg.reward.tracking_lin_vel.weight == pytest.approx(2.2) - assert cfg.env.events.pd_gains is None - - def test_offpolicy_g1_walk_flat_mujoco_td3_uses_td3_task_owner(): cfg = _compose("td3", overrides=["task=g1_walk_flat/mujoco"]) diff --git a/tests/config/test_locomotion_params.py b/tests/config/test_locomotion_params.py index 7a0bc5e94..f3ed8a3c9 100644 --- a/tests/config/test_locomotion_params.py +++ b/tests/config/test_locomotion_params.py @@ -174,18 +174,6 @@ def test_offpolicy_td3_defaults(): assert cfg.algo.algo_params.log_std_min == pytest.approx(-1.6) -def test_offpolicy_td3_g1_task_overrides(): - from hydra import compose, initialize_config_dir - from hydra.core.global_hydra import GlobalHydra - - GlobalHydra.instance().clear() - with initialize_config_dir(config_dir=str(CONF_DIR / "td3"), version_base="1.3"): - cfg = compose("config", overrides=["task=g1_walk_flat/mujoco"]) - assert cfg.training.task_name == "G1WalkFlat" - assert cfg.algo.max_iterations == 100000 - assert cfg.env.actions.joint_pos.scale == pytest.approx(1.0) - - def test_offpolicy_flashsac_g1_task_overrides(): from hydra import compose, initialize_config_dir from hydra.core.global_hydra import GlobalHydra @@ -294,7 +282,6 @@ def test_appo_g1_task_overrides(): assert cfg.algo.max_iterations == 500 assert cfg.algo.save_interval == 100 assert cfg.training.task_name == "G1WalkFlat" - assert "obs_profile" not in cfg.env assert "curriculum" not in cfg.env @@ -315,22 +302,6 @@ def test_ppo_go1_max_iterations(): assert cfg.algo.algorithm.enable_compile is False -def test_ppo_compile_overrides(): - from hydra import compose, initialize_config_dir - from hydra.core.global_hydra import GlobalHydra - - GlobalHydra.instance().clear() - with initialize_config_dir(config_dir=str(CONF_DIR / "ppo"), version_base="1.3"): - cfg = compose( - "config", - overrides=[ - "task=go1_joystick_flat/mujoco", - "algo.algorithm.enable_compile=false", - ], - ) - assert cfg.algo.algorithm.enable_compile is False - - def test_ppo_g1_num_envs(): from hydra import compose, initialize_config_dir from hydra.core.global_hydra import GlobalHydra @@ -341,7 +312,6 @@ def test_ppo_g1_num_envs(): assert cfg.algo.num_envs == 2048 assert cfg.algo.max_iterations == 2200 assert cfg.training.task_name == "G1WalkFlat" - assert "obs_profile" not in cfg.env assert "curriculum" not in cfg.env diff --git a/tests/conftest.py b/tests/conftest.py index c242ad33f..20e4073fe 100644 --- a/tests/conftest.py +++ b/tests/conftest.py @@ -25,8 +25,6 @@ import shutil import pytest -import torch -from uni_rl.ipc.rollout_ring_buffer import RolloutRingBuffer # --------------------------------------------------------------------------- # Dummy flat env — no MuJoCo required @@ -40,9 +38,6 @@ DUMMY_ENV_NAME as _DUMMY_ENV_NAME, ) -_DUMMY_OBS_DIM = 8 -_DUMMY_ACT_DIM = 3 - # Make the dummy env discoverable inside spawn collector subprocesses. _existing = os.environ.get("UNILAB_EXTRA_REGISTRY_PACKAGES", "") _pkgs = [p.strip() for p in _existing.split(",") if p.strip()] @@ -91,36 +86,6 @@ def _isolate_training_logs_for_tests(tmp_path_factory: pytest.TempPathFactory): shutil.rmtree(log_root, ignore_errors=True) -@pytest.fixture -def mp_ctx(): - return torch.multiprocessing.get_context("spawn") - - -@pytest.fixture -def tiny_storage(): - storage = RolloutRingBuffer( - num_envs=4, - num_steps=10, - obs_dim=_DUMMY_OBS_DIM, - action_dim=_DUMMY_ACT_DIM, - num_slots=2, - create=True, - ) - yield storage - storage.cleanup() - - -@pytest.fixture -def tiny_weight_shapes(): - """Small MLP param shapes dict — linear(8,16) + bias, linear(16,3) + bias.""" - return { - "layer1.weight": torch.Size([16, 8]), - "layer1.bias": torch.Size([16]), - "layer2.weight": torch.Size([3, 16]), - "layer2.bias": torch.Size([3]), - } - - @pytest.fixture def mock_env_name() -> str: return _DUMMY_ENV_NAME diff --git a/tests/envs/locomotion/a2/test_a2_joystick_contract.py b/tests/envs/locomotion/a2/test_a2_joystick_contract.py index dfcc3664c..77e4d940a 100644 --- a/tests/envs/locomotion/a2/test_a2_joystick_contract.py +++ b/tests/envs/locomotion/a2/test_a2_joystick_contract.py @@ -2,7 +2,6 @@ from __future__ import annotations -import importlib from collections.abc import Mapping, Sequence from dataclasses import fields, is_dataclass from pathlib import Path @@ -269,23 +268,12 @@ def test_a2_owner_declares_all_randomization_as_manager_events() -> None: assert push.is_global_time is True -def test_a2_registry_has_no_legacy_config_or_runtime_fallback() -> None: +def test_a2_registry_is_manager_only() -> None: registry.ensure_registries() - module = importlib.import_module("unilab.tasks.locomotion.a2.joystick") - assert not hasattr(module, "A2JoystickCfg") - assert not hasattr(module, "A2JoystickFlatEnv") - assert not hasattr(module, "A2JoystickDomainRandomizationProvider") assert registry.list_registered_envs()["A2JoystickFlat"] == { "config_factory": "ManagerBasedRlEnvCfg", "available_backends": ["mujoco"], } - for legacy_override in ( - {"reward_config": {}}, - {"domain_rand": {"randomize_kp": True}}, - {"control_config": {"action_scale": 0.4}}, - ): - with pytest.raises(ValueError, match="has no attribute"): - apply_cfg_overrides(ManagerBasedRlEnvCfg(), legacy_override) def test_a2_registry_executes_real_manager_runtime() -> None: diff --git a/tests/envs/locomotion/g1/test_g1_owner_contract.py b/tests/envs/locomotion/g1/test_g1_owner_contract.py index eb0e1568b..86c3bb6f3 100644 --- a/tests/envs/locomotion/g1/test_g1_owner_contract.py +++ b/tests/envs/locomotion/g1/test_g1_owner_contract.py @@ -591,10 +591,22 @@ def test_g1_owner_materializes_complete_plain_manager_cfg( assert hydra_cfg.training.play_render_mode == "auto" if backend == "isaacgym": assert env_cfg.isaacgym_device_id == 0 + # The subprocess backend consumes the self-contained MJCF scene + # directly; scene fragments and generated terrain stay unset. + assert env_cfg.scene.fragment_files == [] + assert env_cfg.scene.terrain is None + # Effort-mode dofs carry no PD gains, so the owner disables kp/kd + # randomization like the mjwarp/motrix owners. + assert env_cfg.events["pd_gains"] is None # Native rendering (viewer + camera-sensor record) is supported; # playback stays on the base config's auto mode. assert hydra_cfg.training.play_render_mode == "auto" if backend == "genesis": + assert env_cfg.genesis_device_id == 0 + # The in-process backend consumes the self-contained MJCF scene + # directly; scene fragments and generated terrain stay unset. + assert env_cfg.scene.fragment_files == [] + assert env_cfg.scene.terrain is None # Re-declares the MJCF