Adapt dexsim v0.4.0

[yuecideng]

Description

Type of change

• Bug fix (non-breaking change which fixes an issue)
• Enhancement (non-breaking change which improves an existing functionality)
• New feature (non-breaking change which adds functionality)
• Breaking change (existing functionality will not work without user modification)
• Documentation update

Screenshots

Please attach before and after screenshots of the change if applicable.

Checklist

• I have run the black . command to format the code base.
• I have made corresponding changes to the documentation
• I have added tests that prove my fix is effective or that my feature works
• Dependencies have been updated, if applicable.

[Copilot]

Pull request overview

This PR adapts the codebase to dexsim v0.4.0 by replacing the old ray-tracing toggle (enable_rt) with a structured renderer configuration (RenderCfg / --renderer), updating simulation initialization paths, and aligning tests/scripts/docs with the new API.

Changes:

• Introduces RenderCfg and plumbs render_cfg through SimulationManagerCfg, plus a pytest --renderer override.
• Updates many tutorials/examples/tests to use --renderer / RenderCfg and adjusts expected sensor output shapes (e.g., normals/positions).
• Adds rigid-body acceleration accessors (lin_acc, ang_acc, acc) and related documentation.

Reviewed changes

Copilot reviewed 105 out of 105 changed files in this pull request and generated 13 comments.

Show a summary per file

File	Description
tests/sim/solvers/test_srs_solver.py	Imports updated to include RenderCfg (renderer migration).
tests/sim/solvers/test_pytorch_solver.py	Imports updated to include RenderCfg (renderer migration).
tests/sim/solvers/test_pinocchio_solver.py	Imports updated to include RenderCfg (renderer migration).
tests/sim/solvers/test_pink_solver.py	Imports updated to include RenderCfg (renderer migration).
tests/sim/solvers/test_opw_solver.py	Adds RenderCfg import for new renderer configuration path.
tests/sim/solvers/test_differential_solver.py	Imports updated to include RenderCfg (renderer migration).
tests/sim/sensors/test_stereo.py	Removes enable_rt usage; updates expected output channel shapes.
tests/sim/sensors/test_contact.py	Removes enable_rt usage; adds RenderCfg import.
tests/sim/sensors/test_camera.py	Removes enable_rt usage; updates expected output channel shapes.
tests/sim/planners/test_toppra_planner.py	Adds RenderCfg import (renderer migration).
tests/sim/planners/test_motion_generator.py	Adds RenderCfg import (renderer migration).
tests/sim/objects/test_usd.py	Removes enable_rt usage; adds RenderCfg import.
tests/sim/objects/test_soft_object.py	Removes enable_rt usage; adds RenderCfg import.
tests/sim/objects/test_robot.py	Whitespace cleanup.
tests/sim/objects/test_rigid_object.py	Adds rigid-body acceleration tests and RenderCfg import.
tests/sim/objects/test_cloth_object.py	Removes enable_rt usage.
tests/gym/envs/test_embodied_env.py	Removes enable_rt usage; adds RenderCfg import.
tests/gym/envs/managers/test_dataset_functors.py	Whitespace cleanup.
tests/conftest.py	Adds pytest --renderer option and sets global default renderer for tests.
tests/common.py	Whitespace cleanup.
tests/agents/test_shared_rollout.py	Switches env config to render_cfg=RenderCfg(...).
scripts/tutorials/sim/import_usd.py	Migrates CLI to shared launcher args; uses RenderCfg.
scripts/tutorials/sim/gizmo_robot.py	Migrates CLI to shared launcher args; uses RenderCfg.
scripts/tutorials/sim/export_usd.py	Migrates CLI to shared launcher args; uses RenderCfg and updates lighting condition.
scripts/tutorials/sim/create_softbody.py	Migrates CLI to shared launcher args; uses RenderCfg.
scripts/tutorials/sim/create_sensor.py	Migrates CLI to shared launcher args; uses RenderCfg.
scripts/tutorials/sim/create_scene.py	Migrates CLI to shared launcher args; uses RenderCfg.
scripts/tutorials/sim/create_robot.py	Migrates CLI to shared launcher args; uses RenderCfg.
scripts/tutorials/sim/create_rigid_object_group.py	Migrates CLI to shared launcher args; uses RenderCfg.
scripts/tutorials/sim/create_cloth.py	Migrates CLI to shared launcher args; uses RenderCfg; tweaks cloth physical params.
scripts/tutorials/gym/random_reach.py	Adds renderer parameter and passes RenderCfg into sim cfg.
scripts/tutorials/gym/modular_env.py	Migrates CLI to shared launcher args; uses RenderCfg.
scripts/tutorials/grasp/grasp_generator.py	Migrates CLI to shared launcher args; uses RenderCfg.
examples/sim/utility/workspace_analyzer/analyze_plane_workspace.py	Sets explicit RenderCfg(renderer=\"legacy\") for visualization script.
examples/sim/utility/workspace_analyzer/analyze_joint_workspace.py	Sets explicit RenderCfg(renderer=\"legacy\") for visualization script.
examples/sim/utility/workspace_analyzer/analyze_cartesian_workspace.py	Sets explicit RenderCfg(renderer=\"legacy\") for visualization script.
examples/sim/sensors/create_contact_sensor.py	Migrates CLI to shared launcher args; uses RenderCfg.
examples/sim/sensors/batch_camera.py	Migrates CLI to shared launcher args; uses RenderCfg.
examples/sim/scene/scene_demo.py	Migrates CLI to shared launcher args; uses RenderCfg.
examples/sim/gizmo/gizmo_w1.py	Migrates CLI to shared launcher args; uses RenderCfg.
examples/sim/gizmo/gizmo_scene.py	Migrates CLI to shared launcher args; uses RenderCfg.
examples/sim/gizmo/gizmo_robot.py	Migrates CLI to shared launcher args; uses RenderCfg.
examples/sim/gizmo/gizmo_object.py	Migrates CLI to shared launcher args; uses RenderCfg.
examples/sim/gizmo/gizmo_camera.py	Migrates CLI to shared launcher args; uses RenderCfg.
examples/sim/demo/scoop_ice.py	Migrates CLI to shared launcher args; uses RenderCfg.
examples/sim/demo/press_softbody.py	Migrates CLI to shared launcher args; uses RenderCfg.
examples/sim/demo/pick_up_cloth.py	Migrates CLI to shared launcher args; uses RenderCfg; tweaks object params.
examples/sim/demo/grasp_cup_to_caffe.py	Migrates CLI to shared launcher args; uses RenderCfg; adds a sim.update(step=1).
examples/agents/datasets/online_dataset_demo.py	Attempts to migrate from enable_rt to renderer config in gym_config.
embodichain/utils/warp/kinematics/opw_solver.py	Whitespace cleanup.
embodichain/utils/configclass.py	Whitespace cleanup.
embodichain/utils/init.py	Whitespace cleanup.
embodichain/toolkits/urdf_assembly/component.py	Whitespace cleanup.
embodichain/lab/sim/utility/workspace_analyzer/visualizers/voxel_visualizer.py	Whitespace cleanup.
embodichain/lab/sim/utility/workspace_analyzer/visualizers/sphere_visualizer.py	Whitespace cleanup.
embodichain/lab/sim/utility/workspace_analyzer/visualizers/base_visualizer.py	Whitespace cleanup.
embodichain/lab/sim/utility/workspace_analyzer/samplers/base_sampler.py	Whitespace cleanup.
embodichain/lab/sim/utility/workspace_analyzer/constraints/workspace_constraint.py	Whitespace cleanup.
embodichain/lab/sim/utility/workspace_analyzer/constraints/base_constraint.py	Whitespace cleanup.
embodichain/lab/sim/utility/workspace_analyzer/configs/init.py	Whitespace cleanup.
embodichain/lab/sim/utility/workspace_analyzer/caches/cache_manager.py	Whitespace cleanup.
embodichain/lab/sim/utility/workspace_analyzer/caches/base_cache.py	Whitespace cleanup.
embodichain/lab/sim/utility/sim_utils.py	Updates RT detection logic; adjusts dexsim type imports.
embodichain/lab/sim/types.py	Whitespace cleanup.
embodichain/lab/sim/solvers/pinocchio_solver.py	Whitespace cleanup.
embodichain/lab/sim/solvers/opw_solver.py	Whitespace cleanup.
embodichain/lab/sim/solvers/differential_solver.py	Whitespace cleanup.
embodichain/lab/sim/sim_manager.py	Replaces enable_rt with render_cfg; adjusts GPU sync/initialization behavior.
embodichain/lab/sim/robots/dexforce_w1/utils.py	Whitespace cleanup.
embodichain/lab/sim/robots/cobotmagic.py	Updates standalone demo to use RenderCfg and new defaults.
embodichain/lab/sim/objects/rigid_object.py	Adds acceleration properties (lin_acc, ang_acc, acc).
embodichain/lab/sim/objects/gizmo.py	Removes try/except around gizmo callback registration (assumes new API).
embodichain/lab/sim/objects/articulation.py	Minor whitespace cleanup.
embodichain/lab/sim/cfg.py	Introduces RenderCfg and global DEFAULT_RENDERER; tweaks physics defaults.
embodichain/lab/sim/atom_actions.py	Whitespace cleanup.
embodichain/lab/scripts/run_agent.py	Whitespace cleanup.
embodichain/lab/scripts/preview_asset.py	Migrates CLI to --renderer and sets emission lighting defaults.
embodichain/lab/gym/utils/gym_utils.py	Replaces --enable_rt with --renderer; plumbs into env cfg building.
embodichain/lab/gym/envs/tasks/tableware/pour_water/action_bank.py	Whitespace cleanup.
embodichain/lab/gym/envs/managers/randomization/spatial.py	Whitespace cleanup.
embodichain/lab/gym/envs/managers/randomization/physics.py	Whitespace cleanup.
embodichain/lab/gym/envs/embodied_env.py	Whitespace cleanup.
embodichain/lab/gym/envs/action_bank/utils.py	Whitespace cleanup.
embodichain/data/assets/scene_assets.py	Whitespace cleanup.
embodichain/data/assets/robot_assets.py	Whitespace cleanup.
embodichain/data/assets/obj_assets.py	Whitespace cleanup.
embodichain/data/assets/materials.py	Whitespace cleanup.
embodichain/data/assets/eef_assets.py	Whitespace cleanup.
embodichain/agents/rl/train.py	Migrates trainer config from enable_rt to renderer / RenderCfg.
embodichain/agents/rl/models/mlp.py	Whitespace cleanup.
embodichain/agents/engine/data.py	Migrates sim worker cfg from enable_rt to render_cfg=RenderCfg(...).
embodichain/agents/datasets/sampler.py	Whitespace cleanup.
embodichain/agents/datasets/online_data.py	Whitespace cleanup.
docs/sync_readme.py	Formatting/whitespace tweak.
docs/source/tutorial/sensor.rst	Updates tutorial to reference --renderer (replacing --enable_rt).
docs/source/tutorial/robot.rst	Updates tutorial to reference --renderer (replacing --enable_rt).
docs/source/tutorial/gizmo.rst	Updates tutorial to reference --renderer (replacing --enable_rt).
docs/source/overview/sim/sim_rigid_object.md	Documents new rigid-body acceleration accessors.
docs/source/overview/sim/sim_manager.md	Updates config docs to use render_cfg / RenderCfg.
docs/source/guides/cli.md	Updates CLI docs from --enable_rt to --renderer.
docs/source/features/toolkits/grasp_generator.rst	Updates docs to mention renderer backend selection.
docs/source/features/interaction/preview_asset.md	Updates docs to use --renderer.
configs/agents/rl/push_cube/train_config.json	Replaces enable_rt with renderer.
configs/agents/rl/basic/cart_pole/train_config.json	Replaces enable_rt with renderer.
configs/agents/rl/basic/cart_pole/train_config_grpo.json	Replaces enable_rt with renderer.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

SDFConfig is commented out in the dexsim.types import, but it is still used later (e.g., when cfg.sdf_resolution > 0, sdf_cfg = SDFConfig()). This will raise a NameError at runtime. Either restore the SDFConfig import (if still available in dexsim v0.4.0) or update the SDF code path to use the new dexsim API / remove the feature when unsupported.

[Copilot]

gym_config.get("renderer", "legacy") = True is invalid Python (cannot assign to a function call) and will crash before the demo runs. This likely intended to set a key, e.g. gym_config["renderer"] = "legacy" (or another backend string), instead of using the old enable_rt boolean.

[Copilot]

The example command --renderer is missing the required value; the CLI expects one of legacy, hybrid, or fast-rt. As written, python ... --renderer will fail argument parsing. Update the docs to show an explicit backend (e.g., --renderer fast-rt or --renderer hybrid) and clarify that legacy disables ray tracing.

Suggested change

	python scripts/tutorials/sim/create_sensor.py --renderer
	python scripts/tutorials/sim/create_sensor.py --renderer fast-rt
	# Renderer backends: legacy (disables ray tracing), hybrid, fast-rt

[Copilot]

The example python scripts/tutorials/sim/create_robot.py --renderer is missing the renderer backend value. Since --renderer takes an argument (legacy/hybrid/fast-rt), this command will error; please update the docs to include a value (e.g., --renderer fast-rt) and describe which values enable ray tracing.

Suggested change

	python scripts/tutorials/sim/create_robot.py --renderer
	python scripts/tutorials/sim/create_robot.py --renderer fast-rt
	Valid ``--renderer`` values are ``legacy``, ``hybrid``, and ``fast-rt``.
	The ``hybrid`` and ``fast-rt`` backends enable ray tracing; ``legacy`` does not.

[Copilot]

The CLI option description --renderer: Enable ray tracing for better visuals is misleading/incomplete: --renderer requires a value (legacy|hybrid|fast-rt) and only hybrid/fast-rt enable RT features. Update the docs to reflect the accepted values and give an example like --renderer hybrid.

[Copilot]

TestStereoCameraRaster* vs TestStereoCameraFastRT* now run with identical simulation settings (no render_cfg passed), so the suite no longer actually tests different renderer backends and just repeats the same test work. Consider plumbing a renderer argument through setup_simulation and setting render_cfg=RenderCfg(renderer=...) for each variant (or drop the duplicated variants).

[Copilot]

TestCPURT is now identical to TestCPU (no renderer difference is configured after removing enable_rt), and RenderCfg is imported but unused. Either set cfg.sim_cfg.render_cfg=RenderCfg(renderer=...) for the RT variant (and/or rename it), or remove the redundant class so the test intent matches behavior.

[Copilot]

SimulationManagerCfg is created with headless=True unconditionally, even though the script now parses --headless via add_env_launcher_args_to_parser and later conditionally calls sim.open_window(). This can lead to inconsistent window initialization behavior; prefer wiring headless=args.headless into SimulationManagerCfg so CLI flags are respected end-to-end.

[Copilot]

SimulationManagerCfg sets headless=True unconditionally even though --headless is parsed and later used to decide whether to open a window. To make the CLI behave as documented, pass headless=args.headless when constructing SimulationManagerCfg.

[Copilot]

RenderCfg(renderer="fast-rt" if False else "legacy") is effectively always legacy and reads like leftover debugging code. Consider replacing it with a direct literal (or parameterizing via a constant/fixture) to keep the test intent clear.

[Copilot]

Pull request overview

Copilot reviewed 105 out of 105 changed files in this pull request and generated 9 comments.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

In distributed runs, gym_env_cfg.sim_cfg.gpu_id is set from the resolved CUDA device index earlier, but this later assignment overwrites it with the trainer config’s gpu_id. That can select the wrong GPU per-rank. Consider only applying the trainer gpu_id override when not distributed (or set it to the resolved gpu_index/local_rank).

[Copilot]

TestContactFastRT no longer configures a different renderer than the raster test (same setup_simulation("cpu")), so it duplicates test execution under a misleading name. Consider parameterizing by renderer (legacy/hybrid/fast-rt) or removing these classes.

[Copilot]

TestCameraFastRT is currently identical to TestCameraRaster (same setup_simulation("cpu")) and does not set a FastRT renderer anymore, so it duplicates test execution under a misleading name. Either configure render_cfg explicitly for these classes or remove/rename them.

[Copilot]

The docs show --renderer as a flag, but the CLI expects a value (e.g., --renderer legacy|hybrid|fast-rt). Update the example invocation to include a renderer choice so users can copy/paste it successfully.

Suggested change

	python scripts/tutorials/sim/create_sensor.py --renderer
	python scripts/tutorials/sim/create_sensor.py --renderer fast-rt

[Copilot]

The docs show --renderer without an argument, but the CLI requires a backend value (e.g., --renderer legacy|hybrid|fast-rt). Update the command example accordingly.

Suggested change

	python scripts/tutorials/sim/create_robot.py --renderer
	python scripts/tutorials/sim/create_robot.py --renderer fast-rt

[Copilot]

--renderer is documented like a boolean switch here, but it takes a value (legacy/hybrid/fast-rt). Consider documenting it as --renderer <legacy|hybrid|fast-rt> and explaining what each option does.

Suggested change

	- ``--renderer``: Enable ray tracing for better visuals
	- ``--renderer <legacy|hybrid|fast-rt>``: Choose the rendering backend. ``legacy`` uses the standard rasterized renderer, ``hybrid`` uses a hybrid renderer that balances performance and visual quality, and ``fast-rt`` enables the fast ray-tracing renderer for improved visuals.

[Copilot]

TestStereoCameraFastRT no longer configures a different renderer than the raster test (same setup_simulation("cpu")), so this duplicates test execution under a misleading name. Consider parameterizing by renderer via SimulationManagerCfg(render_cfg=RenderCfg(...)) or removing these classes.

[Copilot]

Pull request overview

Copilot reviewed 110 out of 110 changed files in this pull request and generated 8 comments.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

The Raster/FastRT subclasses no longer differ (no renderer selection is passed into SimulationManagerCfg), and RenderCfg is imported but unused. Consider wiring render_cfg=RenderCfg(renderer=...) into SimulationManagerCfg so the tests actually exercise the intended backends and the imports/class names remain meaningful.

[Copilot]

RenderCfg.to_dexsim_flags() logs an error for an unknown renderer but returns None, which can propagate into world_config.renderer and fail later in hard-to-debug ways. Prefer raising a ValueError (or returning a safe default) to enforce invalid values are handled immediately.

[Copilot]

gym_env_cfg.sim_cfg.gpu_id is first set from the resolved CUDA device index (via gpu_index), but then immediately overwritten with the trainer config's gpu_id. This can break multi-GPU/distributed runs where each rank must target a distinct GPU. Consider only setting gpu_id once (derived from device/local_rank), or make the override conditional so it can’t undo the device-based selection.

[Copilot]

DEFAULT_RENDERER is set to "hybrid", but the RenderCfg docstring says "legacy" is the default for backward compatibility. Either change DEFAULT_RENDERER to "legacy" or update the docstring (and any other docs) so the stated default matches runtime behavior.

[Copilot]

The add_env_launcher_args_to_parser docstring says the default renderer is 'legacy', but the argparse option sets default="hybrid". Please align the docstring with the actual default (or change the default to match the documented behavior).

[Copilot]

_build_env_cfg() no longer accepts enable_rt, but it also never sets sim_cfg.render_cfg from config/CLI. This changes behavior vs. previous defaults (often raster/legacy) and makes the renderer impossible to control from the benchmark runtime. Consider threading a renderer argument through and setting gym_env_cfg.sim_cfg.render_cfg = RenderCfg(renderer=renderer) (or reading it from the loaded config).

[Copilot]

Pytest --renderer validation rejects "rt", but the rest of the codebase/CLI accepts it (e.g., add_env_launcher_args_to_parser(... choices=[..., "rt"])). Consider adding "rt" here (and updating the help/error text) so tests can be run against the offline RT backend too.

[Copilot]

SimulationManagerCfg is created with headless=True unconditionally, but later logic uses args.headless to decide whether to open a window. This is inconsistent (the manager may never create a window even when --headless is false). Pass headless=args.headless into SimulationManagerCfg so the CLI flag actually takes effect.

[Copilot]

Pull request overview

Copilot reviewed 111 out of 111 changed files in this pull request and generated 1 comment.

Comments suppressed due to low confidence (1)

embodichain/lab/gym/utils/gym_utils.py:862

• --gym_config is now required=False with default "", but build_env_cfg_from_args() unconditionally calls load_json(args.gym_config), which will raise FileNotFoundError when the flag is omitted. Either make --gym_config required again, or add a real default/guard (e.g., error with a clear message when empty).

    parser.add_argument(
        "--gym_config",
        type=str,
        help="Path to gym config file.",
        default="",
        required=False,
    )
    parser.add_argument(
        "--action_config", type=str, help="Path to action config file.", default=None
    )
    parser.add_argument(
        "--preview",
        help="Whether to preview the environment after launching.",
        default=False,
        action="store_true",
    )
    parser.add_argument(
        "--filter_visual_rand",
        help="Whether to filter out visual randomization.",
        default=False,
        action="store_true",
    )
    parser.add_argument(
        "--filter_dataset_saving",
        help="Whether to filter out dataset saving.",
        default=False,
        action="store_true",
    )


def merge_args_with_gym_config(args: argparse.Namespace, gym_config: dict) -> dict:
    """Merge command-line arguments with gym configuration.

    Command-line arguments will override the corresponding values in the gym configuration.

    Args:
        args (argparse.Namespace): The parsed command-line arguments.
        gym_config (dict): The original gym configuration dictionary.

    Returns:
        dict: The merged gym configuration dictionary.
    """
    merged_config = deepcopy(gym_config)
    merged_config["num_envs"] = args.num_envs
    merged_config["device"] = args.device
    merged_config["headless"] = args.headless
    merged_config["renderer"] = args.renderer
    merged_config["gpu_id"] = args.gpu_id
    merged_config["arena_space"] = args.arena_space
    return merged_config


def build_env_cfg_from_args(
    args: argparse.Namespace,
) -> tuple["EmbodiedEnvCfg", dict, dict]:
    """Build environment configuration from command-line arguments.

    Args:
        args (argparse.Namespace): The parsed command-line arguments.

    Returns:
        tuple[EmbodiedEnvCfg, dict, dict]: A tuple containing the environment configuration object,
            the original gym configuration dictionary, and the action configuration dictionary.
    """
    from embodichain.utils.utility import load_json
    from embodichain.lab.gym.envs import EmbodiedEnvCfg
    from embodichain.lab.sim import SimulationManagerCfg
    from embodichain.lab.sim.cfg import RenderCfg

    gym_config = load_json(args.gym_config)
    gym_config = merge_args_with_gym_config(args, gym_config)

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

The --body_type argument now defaults to "dynamic", but the help string still says "(default: kinematic)". Update the help text (or revert the default) so CLI usage information is accurate.

Suggested change

	help="Body type for rigid objects (default: kinematic).",
	help="Body type for rigid objects (default: dynamic).",

[Copilot]

Pull request overview

Copilot reviewed 110 out of 110 changed files in this pull request and generated 5 comments.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

DEFAULT_RENDERER is set to "hybrid", but the RenderCfg docstring says "legacy" is the default for backward compatibility. Please align the code and docstring (either change the default constant or update the documentation to match the actual default).

[Copilot]

The docs show --renderer being used as a flag, but the CLI expects a value (e.g., --renderer hybrid or --renderer fast-rt). As written, this example command will fail argument parsing; please include an explicit renderer argument in the example.

Suggested change

	python scripts/tutorials/sim/create_robot.py --renderer
	python scripts/tutorials/sim/create_robot.py --renderer hybrid

[Copilot]

The call to _register_default_window_control() is now commented out during initialization. Since there are no remaining call sites, default window interactions (selection/raycast helper) appear to be disabled even when a window is created. If this was done to work around an engine issue, consider guarding it behind a config flag or restricting it to the affected renderer mode instead of disabling it globally.

[Copilot]

--gym_config is now marked required=False, but build_env_cfg_from_args() unconditionally calls load_json(args.gym_config). This means run-env/run-agent can now parse successfully without --gym_config and then fail later with a confusing file error. Consider keeping --gym_config required for those CLIs, or add an explicit validation/error message in build_env_cfg_from_args when it is missing/empty.