### Install and Launch MuJoCo Viewer

Source: https://github.com/project-instinct/instinctlab/blob/main/source/instinctlab/instinctlab/assets/resources/unitree_g1/README.md

Installs the MuJoCo Python bindings and launches the MuJoCo viewer. Ensure you have Python and pip installed.

```bash
pip install mujoco
python -m mujoco.viewer
```

--------------------------------

### Multi Reward Configuration Example

Source: https://github.com/project-instinct/instinctlab/blob/main/DOCS.md

This configuration defines multiple reward terms and groups them for use with the MultiRewardManager. It includes imitation rewards, regularization rewards, and demonstrates how to structure reward groups for single or multi-critic RL setups.

```python
from instinctlab.managers import MultiRewardCfg
from isaaclab.managers import RewardTermCfg as RewTermCfg
from isaaclab.managers import SceneEntityCfg
from isaaclab.utils import configclass
import instinctlab.envs.mdp as instinct_mdp
import isaaclab.envs.mdp as mdp

# Define individual reward terms
@configclass
class RewardsCfg:
    # Imitation rewards (motion matching)
    base_position_imitation_gauss = RewTermCfg(
        func=instinct_mdp.base_position_imitation_gauss,
        weight=0.5,
        params={"std": 0.3},
    )
    base_rot_imitation_gauss = RewTermCfg(
        func=instinct_mdp.base_rot_imitation_gauss,
        weight=0.5,
        params={"std": 0.4, "difference_type": "axis_angle"},
    )
    link_pos_imitation_gauss = RewTermCfg(
        func=instinct_mdp.link_pos_imitation_gauss,
        weight=1.0,
        params={
            "combine_method": "mean_prod",
            "in_base_frame": False,
            "in_relative_world_frame": True,
            "std": 0.3,
        },
    )

    # Regularization rewards
    action_rate_l2 = RewTermCfg(func=mdp.action_rate_l2, weight=-0.1)
    joint_limit = RewTermCfg(
        func=mdp.joint_pos_limits,
        weight=-10.0,
        params={"asset_cfg": SceneEntityCfg("robot", joint_names=[".*"])},
    )
    undesired_contacts = RewTermCfg(
        func=mdp.undesired_contacts,
        weight=-0.1,
        params={
            "sensor_cfg": SceneEntityCfg("contact_forces", body_names=[".*"]),
            "threshold": 1.0,
        },
    )

# Define reward groups (must inherit from MultiRewardCfg)
@configclass
class RewardGroupsCfg(MultiRewardCfg):
    # Single reward group containing all terms
    rewards = RewardsCfg()

    # Optional: Define multiple groups for multi-critic RL
    # rewards_group_1 = RewardsCfg()  # First critic
    # rewards_group_2 = RewardsCfg()  # Second critic

# In your environment configuration
@configclass
class EnvCfg(InstinctLabRLEnvCfg):
    rewards: RewardGroupsCfg = RewardGroupsCfg()
    # ... other configs ...

```

--------------------------------

### Install pre-commit

Source: https://github.com/project-instinct/instinctlab/blob/main/README.md

Install the pre-commit package using pip. This is the first step to enable automatic code formatting.

```bash
pip install pre-commit
```

--------------------------------

### Training Configuration Setup

Source: https://github.com/project-instinct/instinctlab/blob/main/source/instinctlab/instinctlab/tasks/shadowing/beyondmimic/README.md

Imports and instantiates environment and agent configurations for training BeyondMimic policies.

```python
from instinctlab.tasks.shadowing.beyondmimic.config.g1.beyondmimic_plane_cfg import G1BeyondMimicPlaneEnvCfg
from instinctlab.tasks.shadowing.beyondmimic.config.g1.agents.beyondmimic_ppo_cfg import G1BeyondMimicPPORunnerCfg

# Create environment configuration
env_cfg = G1BeyondMimicPlaneEnvCfg()

# Create agent configuration
agent_cfg = G1BeyondMimicPPORunnerCfg()
```

--------------------------------

### Install pre-commit hook

Source: https://github.com/project-instinct/instinctlab/blob/main/README.md

Set up pre-commit to automatically run on every commit. This command installs the necessary Git hooks.

```bash
pre-commit install
```

--------------------------------

### Install InstinctLab Library

Source: https://github.com/project-instinct/instinctlab/blob/main/README.md

Install the instinctlab library in editable mode using pip. This command should be run using a Python interpreter that has Isaac Lab installed.

```bash
python -m pip install -e source/instinctlab
```

--------------------------------

### Instantiate Noisy Grouped RayCaster Camera Configuration

Source: https://github.com/project-instinct/instinctlab/blob/main/DOCS.md

Example of how to import and instantiate configuration classes for the Noisy Grouped RayCaster Camera and its base Grouped RayCaster.

```python
from instinctlab.sensors import GroupedRayCasterCfg, NoisedGroupedRayCasterCameraCfg
```

--------------------------------

### Reward Configuration Examples

Source: https://github.com/project-instinct/instinctlab/blob/main/DOCS.md

Defines reward terms for tracking base position, imitating joint positions, and tracking shadowing commands. Ensure necessary imports are present.

```python
from instinctlab.envs.mdp import RewTermCfg
from isaaclab.managers import SceneEntityCfg
import instinctlab.envs.mdp as instinct_mdp


@configclass
class RewardsCfg:
    # Tracking reward - checks timing
    base_position_tracking = RewTermCfg(
        func=instinct_mdp.base_position_tracking_gauss,
        params={
            "asset_cfg": SceneEntityCfg("robot"),
            "reference_cfg": SceneEntityCfg("motion_reference"),
            "check_at_keyframe_threshold": -1,  # Check at expected time
            "tracking_sigma": 0.2,
        },
    )

    # Imitation reward - checks every timestep
    joint_pos_imitation = RewTermCfg(
        func=instinct_mdp.joint_pos_imitation_gauss,
        params={
            "asset_cfg": SceneEntityCfg("robot"),
            "reference_cfg": SceneEntityCfg("motion_reference"),
            "std": 0.7,
            "masked": True,  # Only compute for unmasked joints
        },
    )

    # Shadowing command reward - tracks command only
    track_shadowing_cmd = RewTermCfg(
        func=instinct_mdp.track_pose_ref_shadowing_cmd_gauss,
        params={
            "command_cfg": SceneEntityCfg("pose_ref_command"),
            "asset_cfg": SceneEntityCfg("robot"),
        },
    )
```

--------------------------------

### Install Instinct-RL Library

Source: https://github.com/project-instinct/instinctlab/blob/main/README.md

Install the instinct_rl library in editable mode using pip. This makes the library available for use in your Python projects.

```bash
python -m pip install -e instinct_rl
```

--------------------------------

### CMakeLists.txt for Unitree G1 Description

Source: https://github.com/project-instinct/instinctlab/blob/main/source/instinctlab/instinctlab/assets/resources/unitree_g1/CMakeLists.txt

This snippet defines the CMake build system configuration for the Unitree G1 robot description package. It specifies the minimum CMake version, project name, and finds required ROS packages like catkin and roslaunch. It also includes directives for installing launch files and meshes.

```cmake
cmake_minimum_required(VERSION 2.8.3)
project(g1_description)

find_package(catkin REQUIRED)

catkin_package()

find_package(roslaunch)

foreach(dir launch meshes)
  install(DIRECTORY ${dir}/
          DESTINATION ${CATKIN_PACKAGE_SHARE_DESTINATION}/${dir})
endforeach(dir)
```

--------------------------------

### Clone Instinct-RL Repository

Source: https://github.com/project-instinct/instinctlab/blob/main/README.md

Clone the instinct_rl repository to your local machine. This is a prerequisite for installing and using the instinct_rl library.

```bash
git clone https://github.com/project-instinct/instinct_rl.git
```

--------------------------------

### Register a new RL environment task

Source: https://github.com/project-instinct/instinctlab/blob/main/README.md

Example of registering a new reinforcement learning environment task using gymnasium. It specifies the entry point for the environment and configuration.

```python
import gymnasium as gym
from . import agents
task_entry = "instinctlab.tasks.shadowing.perceptive.config.g1"
gym.register(
    id="Instinct-Perceptive-Shadowing-G1-Play-v0",
    entry_point="instinctlab.envs:InstinctRlEnv",
    disable_env_checker=True,
    kwargs={
        "env_cfg_entry_point": f"{__name__}.perceptive_shadowing_cfg:G1PerceptiveShadowingEnvCfg_PLAY",
        "instinct_rl_cfg_entry_point": f"{agents.__name__}.instinct_rl_ppo_cfg:G1PerceptiveShadowingPPORunnerCfg",
    },
)
```

--------------------------------

### Clone InstinctLab Repository (HTTPS)

Source: https://github.com/project-instinct/instinctlab/blob/main/README.md

Clone the instinctlab repository using HTTPS. Ensure this is done separately from the Isaac Lab installation directory.

```bash
git clone https://github.com/project-instinct/instinctlab.git
```

--------------------------------

### Clone InstinctLab Repository (SSH)

Source: https://github.com/project-instinct/instinctlab/blob/main/README.md

Clone the instinctlab repository using SSH. Ensure this is done separately from the Isaac Lab installation directory.

```bash
git clone git@github.com:project-instinct/instinctlab.git
```

--------------------------------

### Play Perceptive Shadowing Policy (PPO)

Source: https://github.com/project-instinct/instinctlab/blob/main/source/instinctlab/instinctlab/tasks/shadowing/README.md

Use this command to play a trained Perceptive shadowing policy using PPO. A 'load_run' name is required to load a checkpoint.

```bash
# PPO version
python source/instinctlab/instinctlab/tasks/shadowing/play.py --task=Instinct-Perceptive-Shadowing-G1-v0 --load_run=<run_name>
```

--------------------------------

### Run pre-commit on all files

Source: https://github.com/project-instinct/instinctlab/blob/main/README.md

Execute pre-commit to format all files in the repository. This ensures consistent code style across the project.

```bash
pre-commit run --all-files
```

--------------------------------

### Play BeyondMimic Shadowing Policy

Source: https://github.com/project-instinct/instinctlab/blob/main/source/instinctlab/instinctlab/tasks/shadowing/README.md

Use this command to play a trained BeyondMimic shadowing policy. A 'load_run' name is required to load a checkpoint.

```bash
python scripts/instinct_rl/play.py --task=Instinct-BeyondMimic-Plane-G1-v0 --load_run=<run_name>
```

--------------------------------

### Play Whole Body Shadowing Policy

Source: https://github.com/project-instinct/instinctlab/blob/main/source/instinctlab/instinctlab/tasks/shadowing/README.md

Use this command to play a trained Whole Body shadowing policy. A 'load_run' name is required to load a checkpoint.

```bash
python scripts/instinct_rl/play.py --task=Instinct-Shadowing-WholeBody-Plane-G1-v0 --load_run=<run_name>
```

--------------------------------

### Playing Trained Policies

Source: https://github.com/project-instinct/instinctlab/blob/main/source/instinctlab/instinctlab/tasks/shadowing/beyondmimic/README.md

Command to run the play script for trained BeyondMimic policies, specifying a checkpoint path.

```bash
python source/instinctlab/instinctlab/tasks/shadowing/beyondmimic/play.py --checkpoint /path/to/checkpoint.pt
```

--------------------------------

### Train Perceptive Shadowing Policy (PPO)

Source: https://github.com/project-instinct/instinctlab/blob/main/source/instinctlab/instinctlab/tasks/shadowing/README.md

Use this command to train the Perceptive shadowing policy using PPO. Ensure the task ID is correctly specified.

```bash
# PPO version
python scripts/instinct_rl/train.py --headless --task=Instinct-Perceptive-Shadowing-G1-v0
```

--------------------------------

### Play Trained Parkour Policy

Source: https://github.com/project-instinct/instinctlab/blob/main/source/instinctlab/instinctlab/tasks/parkour/README.md

Execute this command to play a trained Parkour policy. A `load_run` name is required to load a checkpoint. Use `--no_resume` to visualize the untrained policy.

```bash
python source/instinctlab/instinctlab/tasks/parkour/scripts/play.py --task=Instinct-Parkour-Target-Amp-G1-v0 --load_run=<run_name>
```

--------------------------------

### Train Parkour Policy

Source: https://github.com/project-instinct/instinctlab/blob/main/source/instinctlab/instinctlab/tasks/parkour/README.md

Use this command to train the Parkour policy. Ensure the task ID is correctly specified in the configuration file.

```bash
python scripts/instinct_rl/train.py --headless --task=Instinct-Parkour-Target-Amp-G1-v0
```

--------------------------------

### Train BeyondMimic Shadowing Policy

Source: https://github.com/project-instinct/instinctlab/blob/main/source/instinctlab/instinctlab/tasks/shadowing/README.md

Use this command to train the BeyondMimic shadowing policy. Ensure the task ID is correctly specified.

```bash
python scripts/instinct_rl/train.py --headless --task=Instinct-BeyondMimic-Plane-G1-v0
```

--------------------------------

### Train RL Task with Instinct-RL

Source: https://github.com/project-instinct/instinctlab/blob/main/README.md

Execute the training script for an Instinct-RL task. The `--task` argument specifies the reinforcement learning task, and `--headless` runs Isaac Sim without a graphical interface.

```bash
python scripts/instinct_rl/train.py --task=Instinct-Shadowing-WholeBody-Plane-G1-Play-v0 --headless
```

--------------------------------

### Register Virtual Obstacles with Sensor

Source: https://github.com/project-instinct/instinctlab/blob/main/DOCS.md

Register virtual obstacles with the sensor during environment initialization to enable penetration detection. This should be called as a startup event.

```python
sensor.register_virtual_obstacles(terrain.virtual_obstacles)
```

```python
register_virtual_obstacles = EventTerm(
        func=instinct_mdp.register_virtual_obstacle_to_sensor,
        mode="startup",
        params={
            "sensor_cfgs": SceneEntityCfg("leg_volume_points"),
        },
    )
```

--------------------------------

### Register Virtual Obstacles with Sensors

Source: https://github.com/project-instinct/instinctlab/blob/main/DOCS.md

Register virtual obstacles to a sensor at environment startup using EventTerm. This allows sensors to detect and report interactions with these obstacles.

```python
from instinctlab.envs.mdp import EventTerm
from instinctlab.envs.mdp.events import register_virtual_obstacle_to_sensor
from isaaclab.managers import SceneEntityCfg

# In your EventsCfg class
class EventsCfg:
    # Register virtual obstacles to volume points sensor at startup
    register_virtual_obstacles = EventTerm(
        func=register_virtual_obstacle_to_sensor,
        mode="startup",  # Register once at environment startup
        params={
            "sensor_cfgs": SceneEntityCfg("leg_volume_points"),
            # Or multiple sensors:
            # "sensor_cfgs": [
            #     SceneEntityCfg("leg_volume_points"),
            #     SceneEntityCfg("arm_volume_points"),
            # ],
        },
    )

```

--------------------------------

### Terrain Importer Configuration with Virtual Obstacles

Source: https://github.com/project-instinct/instinctlab/blob/main/DOCS.md

This snippet shows the import statements for configuring terrain importers and various virtual obstacle types. It sets up the necessary classes for defining terrain generation and virtual obstacle parameters.

```python
from instinctlab.terrains import TerrainImporterCfg
from instinctlab.terrains.virtual_obstacle import (
    PluckerEdgeCylinderCfg,
    RansacEdgeCylinderCfg,
    GreedyconcatEdgeCylinderCfg,
    RayEdgeCylinderCfg,
)
from isaaclab.terrains import TerrainGeneratorCfg
from isaaclab.sensors import patterns
```

--------------------------------

### Train Whole Body Shadowing Policy

Source: https://github.com/project-instinct/instinctlab/blob/main/source/instinctlab/instinctlab/tasks/shadowing/README.md

Use this command to train the Whole Body shadowing policy. Ensure the task ID is correctly specified.

```bash
python scripts/instinct_rl/train.py --headless --task=Instinct-Shadowing-WholeBody-Plane-G1-v0
```

--------------------------------

### Matching Motion Reference with Scene Terrain

Source: https://github.com/project-instinct/instinctlab/blob/main/DOCS.md

Configures an event to match a motion reference with the scene terrain during startup. This is useful for aligning simulated environments with real-world motion data.

```python
from instinctlab.terrains import MotionMatchedTerrainCfg

# Add this to your event cfg
class EventsCfg:
    match_motion_ref_with_scene = EventTermCfg(
        func=instinct_mdp.match_motion_ref_with_scene,
        mode="startup",
        params={
            "motion_ref_cfg": SceneEntityCfg("motion_reference"),
        },
    )
```

--------------------------------

### Terrain Importer Configuration with Virtual Obstacles

Source: https://github.com/project-instinct/instinctlab/blob/main/DOCS.md

Configure the TerrainImporterCfg to include various virtual obstacle detection methods like Plucker, RANSAC, Greedy, and Ray-based edge detection. Each method has specific parameters for thresholding, radius, and resolution.

```python
terrain_cfg = TerrainImporterCfg(
    prim_path="/World/ground",
    terrain_type="generator",  # or "hacked_generator" for custom generator
    terrain_generator=TerrainGeneratorCfg(
        # ... terrain generator config ...
    ),
    virtual_obstacles={
        # Plucker-based edge detection (merges collinear edges)
        "edges_plucker": PluckerEdgeCylinderCfg(
            angle_threshold=70.0,  # degrees
            cylinder_radius=0.2,  # meters
            num_grid_cells=64**3,  # spatial grid resolution
        ),
        # RANSAC-based edge detection (robust to noise)
        "edges_ransac": RansacEdgeCylinderCfg(
            angle_threshold=70.0,
            cylinder_radius=0.2,
            max_iter=500,
            point_distance_threshold=0.04,
            min_points=5,
            cluster_eps=0.08,
        ),
        # Greedy concatenation (connects adjacent edges)
        "edges_greedy": GreedyconcatEdgeCylinderCfg(
            angle_threshold=70.0,
            cylinder_radius=0.2,
            adjacent_angle_threshold=30.0,
            min_points=5,
        ),
        # Ray-based edge detection (from multiple viewpoints)
        "edges_ray": RayEdgeCylinderCfg(
            cylinder_radius=0.2,
            max_iter=500,
            point_distance_threshold=0.005,
            min_points=15,
            cluster_eps=0.08,
            ray_pattern=patterns.GridPatternCfg(
                resolution=0.01,
                size=[6, 6],
                direction=(0.0, 0.0, -1.0),
            ),
            ray_offset_pos=[0.0, 0.0, 1.0],
            max_ray_depth=8.0,
            depth_canny_thresholds=[250, 300],
            normal_canny_thresholds=[80, 250],
            cutoff_z_height=0.1,
        ),
        # Disable a virtual obstacle by setting to None
        # "edges": None,
    },
)
```
```

--------------------------------

### Configure Noisy Camera

Source: https://github.com/project-instinct/instinctlab/blob/main/DOCS.md

Configures a NoisyGroupedRayCasterCameraCfg to simulate a noisy camera, like a RealSense D435i. It includes mesh targets, a detailed offset configuration with trigonometric calculations, and a comprehensive noise pipeline.

```python
noisy_camera = NoisyGroupedRayCasterCameraCfg(
    prim_path="{ENV_REGEX_NS}/Robot/torso_link",
    mesh_prim_paths=[
        "/World/ground",
        # NOTE: Don't forget to add the robot links in robot-specific configuration file.
        # NoisyGroupedRayCasterCameraCfg.RaycastTargetCfg(prim_expr="/World/envs/env_.*/Robot/torso_link/visuals")
    ],
    offset=NoisyGroupedRayCasterCameraCfg.OffsetCfg(
        pos=(
            0.04764571478 + 0.0039635 - 0.0042 * math.cos(math.radians(48)),
            0.015,
            0.46268178553 - 0.044 + 0.0042 * math.sin(math.radians(48)) + 0.016,
        ),
        rot=(
            math.cos(math.radians(0.5) / 2) * math.cos(math.radians(48) / 2),
            math.sin(math.radians(0.5) / 2),
            math.sin(math.radians(48) / 2),
            0.0,
        ),
        convention="world",
    ),
    ray_alignment="yaw",
    pattern_cfg=patterns.PinholeCameraPatternCfg(
        focal_length=1.0,
        horizontal_aperture=2 * math.tan(math.radians(87) / 2),  # fovx
        vertical_aperture=2 * math.tan(math.radians(58) / 2),  # fovy
        height=int(270 / 10),
        width=int(480 / 10),
    ),
    data_types=["distance_to_image_plane"],
    noise_pipeline={
        # "depth_contour_noise": DepthContourNoiseCfg(
        #     contour_threshold=1.8,  # in [m]
        #     maxpool_kernel_size=1,
        # ),
        "depth_artifact_noise": DepthArtifactNoiseCfg(),
        "stereo_noise": RangeBasedGaussianNoiseCfg(
            max_value=1.2,
            min_value=0.12,
            noise_std=0.02,
        ),
        "sky_artifact_noise": DepthSkyArtifactNoiseCfg(),
        # "stereo_too_close_noise": StereoTooCloseNoiseCfg(),
        # These last two noise model will affect the processing on the onboard device.
        "gaussian_blur_noise": GaussianBlurNoiseCfg(
            kernel_size=3,
            sigma=0.5,
        ),
        "normalize": DepthNormalizationCfg(
            depth_range=(0.0, 1.5),
            normalize=True,
        ),
        "crop_and_resize": CropAndResizeCfg(
            crop_region=(2, 2, 2, 2),
            resize_shape=(18, 32),
        ),
    },
    # data_histories={"distance_to_image_plane": 5},
    update_period=1 / 60,
    debug_vis=False,
    depth_clipping_behavior="max",  # clip to the maximum value
    min_distance=0.05,
    max_distance=2.0,
)
```

--------------------------------

### Export Parkour Policy to ONNX

Source: https://github.com/project-instinct/instinctlab/blob/main/source/instinctlab/instinctlab/tasks/parkour/README.md

This command exports the trained Parkour policy to ONNX format for onboard deployment. It requires a `load_run` name and uses `--exportonnx` and `--useonnx` flags.

```bash
python source/instinctlab/instinctlab/tasks/parkour/scripts/play.py --task=Instinct-Parkour-Target-Amp-G1-v0 --load_run=<run_name> --exportonnx --useonnx
```

--------------------------------

### Configure Pylance extraPaths in VSCode

Source: https://github.com/project-instinct/instinctlab/blob/main/README.md

Add the path to your extension repository to VSCode's Pylance settings to resolve indexing issues. This helps Pylance correctly analyze your project's code.

```json
{
    "python.analysis.extraPaths": [
        "<path-to-ext-repo>/source/instinctlab"
    ]
}
```

--------------------------------

### Configure Grouped Ray Caster

Source: https://github.com/project-instinct/instinctlab/blob/main/DOCS.md

Configures a GroupedRayCasterCfg for scene entity querying. It specifies the prim path, offset, ray alignment, pattern, and mesh targets. Useful for general-purpose ray casting in a scene.

```python
grouped_ray_caster = GroupedRayCasterCfg(
    prim_path="{ENV_REGEX_NS}/Robot/torso_link",
    offset=RayCasterCfg.OffsetCfg(pos=(0.0, 0.0, 20.0)),
    ray_alignment="yaw",
    pattern_cfg=patterns.GridPatternCfg(resolution=0.1, size=[1.6, 1.0]),
    debug_vis=False,
    mesh_prim_paths=[
        "/World/ground",
        # NOTE: Don't forget to add the robot links in robot-specific configuration file.
        # GroupedRayCasterCfg.RaycastTargetCfg(prim_expr="/World/envs/env_.*/Robot/torso_link/visuals")
    ],
)
```

--------------------------------

### Base Pose Reference Command Configuration

Source: https://github.com/project-instinct/instinctlab/blob/main/DOCS.md

Defines a base pose reference command, including position and orientation, relative to the robot's anchor frame. It uses axis-angle representation and does not include real-time updates or the current state command.

```python
from isaaclab.envs.mdp import PoseRefCommandCfg, JointPosRefCommandCfg
from isaaclab.managers import SceneEntityCfg

# Base pose reference command (position + orientation)
pose_ref_command = PoseRefCommandCfg(
    motion_reference=SceneEntityCfg("motion_reference"),
    asset_cfg=SceneEntityCfg("robot"),
    anchor_frame="robot",  # Reference in robot's local frame
    rotation_mode="axis_angle",  # Use axis-angle representation
    realtime_mode=0,  # No real-time updates
    current_state_command=False,  # Don't include current state
)
```

--------------------------------

### Export Policy as ONNX Model

Source: https://github.com/project-instinct/instinctlab/blob/main/README.md

Add the `--exportonnx` flag to the `play.py` script to export the trained policy as an ONNX model. This exported model can then be used with the `instinct_onboard` workflow for deployment on a robot.

```bash
python play.py --exportonnx
```

--------------------------------

### Joint Position Reference Command Configuration

Source: https://github.com/project-instinct/instinctlab/blob/main/DOCS.md

Configures a joint position reference command that includes the robot's current joint positions. This is useful for tasks requiring precise joint control based on motion reference data.

```python
from isaaclab.envs.mdp import PoseRefCommandCfg, JointPosRefCommandCfg
from isaaclab.managers import SceneEntityCfg

# Joint position reference command
joint_pos_ref_command = JointPosRefCommandCfg(
    motion_reference=SceneEntityCfg("motion_reference"),
    asset_cfg=SceneEntityCfg("robot"),
    current_state_command=True,  # Include current joint positions
)
```

--------------------------------

### Accessing Virtual Obstacles in Rewards

Source: https://github.com/project-instinct/instinctlab/blob/main/DOCS.md

Define reward terms that penalize penetration into virtual obstacles using volume_points_penetration or volume_points_step_safety functions. These require sensor configurations and optionally contact force sensors.

```python
from instinctlab.envs.mdp import RewTermCfg
import instinctlab.envs.mdp as instinct_mdp

# Reward that penalizes penetration into virtual obstacles
volume_points_penetration = RewTermCfg(
    func=instinct_mdp.volume_points_penetration,
    params={
        "sensor_cfgs": [SceneEntityCfg("leg_volume_points"), SceneEntityCfg("arm_volume_points")],
    },
)

# Reward that penalizes penetration into virtual obstacles
volume_points_step_safety = RewTermCfg(
    func=instinct_mdp.volume_points_step_safety,
    params={
        "sensor_cfgs": [SceneEntityCfg("leg_volume_points"), SceneEntityCfg("arm_volume_points")],
        "contact_forces_cfg": SceneEntityCfg("contact_forces"),
    },
)

```

--------------------------------

### Configure AMASS Motion Data

Source: https://github.com/project-instinct/instinctlab/blob/main/DOCS.md

Configures the motion buffer for the AMASS dataset, specifying the path, buffer length, frame interval, and sampling strategy for motion data.

```python
from instinctlab.motion_reference import MotionReferenceManagerCfg
from instinctlab.motion_reference import AmassMotionCfgBase

# Configure motion buffer (e.g., AMASS dataset)
class AMASSMotionCfg(AmassMotionCfgBase):
    path = "~/Datasets/AMASS_dataset"
    motion_bin_length_s = 1.0
    frame_interval_s = 0.1
    env_starting_stub_sampling_strategy = "concat_motion_bins"

# Configure motion reference manager and put it in SceneCfg
motion_reference_cfg = MotionReferenceManagerCfg(
    prim_path="{ENV_REGEX_NS}/Robot/torso_link",
    robot_model_path="path/to/robot.urdf",
    reference_prim_path="/World/envs/env_.*/RobotReference/torso_link",
    link_of_interests=[
        "pelvis", "torso_link",
        "left_shoulder_roll_link", "right_shoulder_roll_link",
        "left_elbow_link", "right_elbow_link",
        # ... more links
    ],
    frame_interval_s=0.1,  # 10 Hz frame rate
    update_period=0.02,     # Update every 0.02s (50 Hz)
    num_frames=10,          # Look ahead 10 frames (1 second)
    data_start_from="current_time",  # Start from current timestep
    motion_buffers={
        "AMASSMotion": AMASSMotionCfg(),
    },
)
```

--------------------------------

### Termination Term for Motion Reference Exhaustion

Source: https://github.com/project-instinct/instinctlab/blob/main/DOCS.md

Configures a termination term to detect when the motion reference data is exhausted. This is applied per motion reference manager.

```python
from isaaclab.managers import TerminationTermCfg

# Add this to your termination cfg
class TerminationsCfg:
    motion_reference_exhausted = TerminationTermCfg(
        func=instinct_mdp.dataset_exhausted,
        params={
            "reference_cfg": SceneEntityCfg("motion_reference"),
            # "reset_without_notice": True,
            #### If True, this termination term will only reset the exhausted env in the motion reference manager, but not return any True to reset the environment.
        },
    )
```

=== COMPLETE CONTENT === This response contains all available snippets from this library. No additional content exists. Do not make further requests.