Class HalfCheetahEnvV5
- Namespace
- Gym.Environments.Envs.MuJoCo
- Assembly
- Gym.Environments.dll
HalfCheetah-v5: a planar cheetah (a torso and two legs of thigh, shin and foot) driven by 6 joint torques in
[-1, 1]. The reward pays the forward velocity and charges the squared torques; the episode never ends on its own
(make's 1000-step limit truncates it). Byte for byte what Gymnasium's HalfCheetah-v5 computes on MuJoCo
4.0.0 (DR-61, DR-64): observations, rewards, infos and the generator's draws.
[PyName("HalfCheetahEnv")]
public class HalfCheetahEnvV5 : MujocoEnv, IEnv, IDisposable, IDynamicAttributes
- Inheritance
-
HalfCheetahEnvV5
- Implements
- Inherited Members
- Extension Methods
Remarks
Observation (float64): the joint positions without the root's x (unless
exclude_current_positions_from_observation=False), then all joint velocities, 17 values by default.
Reward: forward_reward_weight * x_velocity - ctrl_cost_weight * sum(action**2), the control cost in the
action's dtype (float32 for the action space's samples, so the info's reward_ctrl is an np.float32
0-d array there). Info: x_position, x_velocity, reward_forward, reward_ctrl;
the reset's info holds x_position.
Not thread-safe. Reference: gymnasium/envs/mujoco/half_cheetah_v5.py.
Its info reads typed (DR-91): HalfCheetahInfoV5 info = step.Info; (HalfCheetahInfoV5), with the dict's own values still behind the indexer.
Constructors
HalfCheetahEnvV5(HalfCheetahEnvV5Options)
Builds the instance from its options record: the properties set, the constructor's defaults for the rest.
public HalfCheetahEnvV5(HalfCheetahEnvV5Options options)
Parameters
optionsHalfCheetahEnvV5OptionsThe options.
Exceptions
- ArgumentNullException
Thrown when
optionsis null.
HalfCheetahEnvV5(RenderMode, string, int, IReadOnlyDictionary<string, object?>?, double, double, double, bool, int, int, int?, string?, int, IReadOnlyDictionary<int, bool>?, EnvViewerFactory?)
Builds the env with the render mode given as the RenderMode enum (DR-84, DR-93), as
HalfCheetahEnvV5(string, int, IReadOnlyDictionary<string, object?>?, double, double, double, bool, string?, int, int, int?, string?, int, IReadOnlyDictionary<int, bool>?, EnvViewerFactory?)
builds it from the mode's string: new HalfCheetahEnvV5(RenderMode.RgbArray) builds what
new HalfCheetahEnvV5(renderMode: "rgb_array") builds, the member reaching the env as Gymnasium's string
(Name(RenderMode)), which its RenderMode property then holds.
public HalfCheetahEnvV5(RenderMode renderMode, string xmlFile = "half_cheetah.xml", int frameSkip = 5, IReadOnlyDictionary<string, object?>? defaultCameraConfig = null, double forwardRewardWeight = 1, double ctrlCostWeight = 0.1, double resetNoiseScale = 0.1, bool excludeCurrentPositionsFromObservation = true, int width = 480, int height = 480, int? cameraId = null, string? cameraName = null, int maxGeom = 1000, IReadOnlyDictionary<int, bool>? visualOptions = null, EnvViewerFactory? viewerFactory = null)
Parameters
renderModeRenderModeThe render mode, which the env must list in its
metadata["render_modes"]as for the string; it behaves exactly as its Gymnasium string does in the other constructor.xmlFilestringThe model: a bundled file name (
xml_file, DR-63).frameSkipintPhysics steps per env step (
frame_skip).defaultCameraConfigIReadOnlyDictionary<string, object>The free camera's settings (
default_camera_config); null is Gymnasium's default,{"distance": 4.0}, and an empty dictionary leaves MuJoCo's camera as it is.forwardRewardWeightdoubleThe weight of the forward velocity (
forward_reward_weight).ctrlCostWeightdoubleThe weight of the squared torques (
ctrl_cost_weight).resetNoiseScaledoubleThe scale of the reset's noise (
reset_noise_scale).excludeCurrentPositionsFromObservationboolWhether the root's x leaves the observation (
exclude_current_positions_from_observation).widthintThe rendered frame's width (
width).heightintThe rendered frame's height (
height).cameraIdint?The camera to render from (
camera_id).cameraNamestringThe camera to render from by name (
camera_name).maxGeomintThe most geoms a frame draws (
max_geom).visualOptionsIReadOnlyDictionary<int, bool>MuJoCo visual flags to switch (
visual_options).viewerFactoryEnvViewerFactoryGym.NET's
viewer_factory(DR-20) forhumanmode.
Remarks
The mode is required and comes first, so new HalfCheetahEnvV5() keeps meaning the other constructor, without a
mode; leave the mode out there for no rendering. Every other parameter means what it means there.
Exceptions
- ArgumentOutOfRangeException
Thrown for a value RenderMode does not declare (a cast integer such as
(RenderMode)99).- NotSupportedException
xmlFileis not a bundled model.- OverflowError
A frame skip of 0 (Python's
OverflowErrorderiving the frame rate).- ValueError
Both a camera id and a camera name.
HalfCheetahEnvV5(string, int, IReadOnlyDictionary<string, object?>?, double, double, double, bool, string?, int, int, int?, string?, int, IReadOnlyDictionary<int, bool>?, EnvViewerFactory?)
Creates the env (Gymnasium's HalfCheetahEnv.init): the model, its spaces and the frame rate the time
step gives.
public HalfCheetahEnvV5(string xmlFile = "half_cheetah.xml", int frameSkip = 5, IReadOnlyDictionary<string, object?>? defaultCameraConfig = null, double forwardRewardWeight = 1, double ctrlCostWeight = 0.1, double resetNoiseScale = 0.1, bool excludeCurrentPositionsFromObservation = true, string? renderMode = null, int width = 480, int height = 480, int? cameraId = null, string? cameraName = null, int maxGeom = 1000, IReadOnlyDictionary<int, bool>? visualOptions = null, EnvViewerFactory? viewerFactory = null)
Parameters
xmlFilestringThe model: a bundled file name (
xml_file, DR-63).frameSkipintPhysics steps per env step (
frame_skip).defaultCameraConfigIReadOnlyDictionary<string, object>The free camera's settings (
default_camera_config); null is Gymnasium's default,{"distance": 4.0}, and an empty dictionary leaves MuJoCo's camera as it is.forwardRewardWeightdoubleThe weight of the forward velocity (
forward_reward_weight).ctrlCostWeightdoubleThe weight of the squared torques (
ctrl_cost_weight).resetNoiseScaledoubleThe scale of the reset's noise (
reset_noise_scale).excludeCurrentPositionsFromObservationboolWhether the root's x leaves the observation (
exclude_current_positions_from_observation).renderModestringGymnasium's
render_mode:human,rgb_array,depth_array,rgbd_tupleor null.widthintThe rendered frame's width (
width).heightintThe rendered frame's height (
height).cameraIdint?The camera to render from (
camera_id).cameraNamestringThe camera to render from by name (
camera_name).maxGeomintThe most geoms a frame draws (
max_geom).visualOptionsIReadOnlyDictionary<int, bool>MuJoCo visual flags to switch (
visual_options).viewerFactoryEnvViewerFactoryGym.NET's
viewer_factory(DR-20) forhumanmode.
Exceptions
- NotSupportedException
xmlFileis not a bundled model.- OverflowError
A frame skip of 0 (Python's
OverflowErrorderiving the frame rate).- ValueError
Both a camera id and a camera name.
Properties
ClassMetadata
Gymnasium's class attribute metadata: the four render modes, without a frame rate (each instance derives
its own from the time step). A new dict on every read.
public static Metadata ClassMetadata { get; }
Property Value
CtrlCostWeight
The weight of the squared torques (_ctrl_cost_weight), read every step.
[PyName("_ctrl_cost_weight")]
public double CtrlCostWeight { get; set; }
Property Value
ExcludeCurrentPositionsFromObservation
Whether the root's x leaves the observation (_exclude_current_positions_from_observation). Footgun:
changing it after construction changes the observation's length but not the observation space, as in Gymnasium.
[PyName("_exclude_current_positions_from_observation")]
public bool ExcludeCurrentPositionsFromObservation { get; set; }
Property Value
ForwardRewardWeight
The weight of the forward velocity (_forward_reward_weight), read every step.
[PyName("_forward_reward_weight")]
public double ForwardRewardWeight { get; set; }
Property Value
ObservationStructure
How the observation is laid out (observation_structure): skipped_qpos, qpos and qvel
counts, as Python ints.
public Info ObservationStructure { get; }
Property Value
ResetNoiseScale
The scale of the reset's noise (_reset_noise_scale), read every reset.
[PyName("_reset_noise_scale")]
public double ResetNoiseScale { get; set; }
Property Value
Methods
ControlCost(NDArray)
The control cost of action (control_cost): ctrl_cost_weight * sum(action**2),
computed in the action's dtype and returned widened exactly (Gymnasium returns an np.float32 for a
float32 action).
[PyName("control_cost")]
public double ControlCost(NDArray action)
Parameters
actionNDArrayThe action.
Returns
- double
The cost.
Exceptions
- ArgumentNullException
actionis null.- NotSupportedException
An action of a dtype whose cost NumPy keeps in that dtype (float16, complex).
GetResetInfo()
The reset's info (_get_reset_info): x_position, an np.float64.
protected override Info GetResetInfo()
Returns
- Info
The info.
ResetModel()
The episode's start (reset_model): positions init_qpos + uniform(-scale, scale, nq), then
velocities init_qvel + scale * standard_normal(nv), drawn in that order.
protected override NDArray ResetModel()
Returns
- NDArray
The first observation.
Step(NDArray)
Applies action for FrameSkip physics steps and reports the transition
(step).
public override StepResult<NDArray> Step(NDArray action)
Parameters
actionNDArrayThe 6 torques, float32 (the action space's samples) or any real dtype.
Returns
- StepResult<NDArray>
The observation, the float64 reward,
terminated=False,truncated=Falseand the info.
Exceptions
- ArgumentNullException
actionis null.- ValueError
The action is not of shape
(6,), with Gymnasium's message.- NotSupportedException
An action of a dtype whose cost NumPy keeps in that dtype (float16, complex).