The TriFingerPlatform Class
The TriFingerPlatform class provides access to the
simulation of the TriFinger platform using the same interface as for the real
robot.
For a detailed explanation of the software interface on the real robot, and the logic behind it, please check the paper.
API Documentation
- class trifinger_simulation.TriFingerPlatform(visualization=False, initial_robot_position=None, initial_object_pose=None, enable_cameras=False, time_step_s=0.001, object_type=ObjectType.COLORED_CUBE, camera_delay_steps=90)[source]
Wrapper around the simulation providing the same interface as
robot_interfaces::TriFingerPlatformFrontend.The following methods of the robot_interfaces counterpart are not supported:
get_robot_status()
wait_until_timeindex()
Initialize.
- Parameters:
visualization (bool) – Set to true to run visualization.
initial_robot_position (Sequence[float] | None) – Initial robot joint angles
initial_object_pose – Only if
object_type == COLORED_CUBE: Initial pose for the manipulation object. Can be any object with attributesposition(x, y, z) andorientation(x, y, z, w). If not set, a default pose is used.enable_cameras (bool) – Set to true to enable rendered images in the camera observations. If false camera observations are still available but the images will not be initialized. By default this is disabled as rendering of images takes a lot of computational power. Therefore the cameras should only be enabled if the images are actually used.
time_step_s (float) – Simulation time step duration in seconds.
object_type (ObjectType) – Which type of object to load. This also influences some other aspects: When using the cube, the camera observation will contain an attribute
object_pose.camera_delay_steps (int) – Number of time steps by which camera observations are held back after they are generated. This is used to simulate the delay of the camera observation that is happening on the real system due to processing (mostly the object detection).
- __init__(visualization=False, initial_robot_position=None, initial_object_pose=None, enable_cameras=False, time_step_s=0.001, object_type=ObjectType.COLORED_CUBE, camera_delay_steps=90)[source]
Initialize.
- Parameters:
visualization (bool) – Set to true to run visualization.
initial_robot_position (Sequence[float] | None) – Initial robot joint angles
initial_object_pose – Only if
object_type == COLORED_CUBE: Initial pose for the manipulation object. Can be any object with attributesposition(x, y, z) andorientation(x, y, z, w). If not set, a default pose is used.enable_cameras (bool) – Set to true to enable rendered images in the camera observations. If false camera observations are still available but the images will not be initialized. By default this is disabled as rendering of images takes a lot of computational power. Therefore the cameras should only be enabled if the images are actually used.
time_step_s (float) – Simulation time step duration in seconds.
object_type (ObjectType) – Which type of object to load. This also influences some other aspects: When using the cube, the camera observation will contain an attribute
object_pose.camera_delay_steps (int) – Number of time steps by which camera observations are held back after they are generated. This is used to simulate the delay of the camera observation that is happening on the real system due to processing (mostly the object detection).
- Action(torque=None, position=None)
- get_current_timeindex()
- get_timestamp_ms(t)
- get_desired_action(t)
- get_applied_action(t)
- get_robot_observation(t)
Get observation of the robot state (joint angles, torques, etc.). See
trifinger_simulation.SimFinger.get_observation().
- get_camera_observation(t)[source]
Get camera observation at time step t.
Important
Actual images are only rendered if the class was created with enable_cameras=True in the constructor, otherwise the images in the observation are set to None. Other fields are still valid, though.
- Parameters:
t (int) – The time index of the step for which the observation is requested. Only the value returned by the last call of
append_desired_action()is valid.- Returns:
Observations of the three cameras. Depending of the object type used, this may also contain the object pose.
- Raises:
ValueError – If invalid time index
tis passed.- Return type:
- class trifinger_simulation.ObjectPose[source]
A pure-python copy of trifinger_object_tracking::ObjectPose.
- confidence
Estimate of the confidence for this pose observation.
- Type:
float
- orientation
Orientation of the object as (x, y, z, w) quaternion.
- Type:
array
- position
Position (x, y, z) of the object. Units are meters.
- Type:
array
- class trifinger_simulation.CameraObservation[source]
Pure-python copy of trifinger_cameras.camera.CameraObservation.
- image
The image.
- Type:
array
- timestamp
Timestamp when the image was received.
- Type:
float
- class trifinger_simulation.TriCameraObservation[source]
Python version of trifinger_cameras::TriCameraObservation.
This is a pure-python implementation of trifinger_cameras::TriCameraObservation, so we don’t need to depend on trifinger_cameras here.
- cameras
List of observations of cameras “camera60”, “camera180” and “camera300” (in this order).
- Type:
list of
CameraObservation
- class trifinger_simulation.TriCameraObjectObservation[source]
Python version of trifinger_object_tracking::TriCameraObjectObservation.
This is a pure-python implementation of trifinger_object_tracking::TriCameraObjectObservation, so we don’t need to depend on trifinger_object_tracking here.
- cameras
List of observations of cameras “camera60”, “camera180” and “camera300” (in this order).
- Type:
list of
CameraObservation
- filtered_object_pose
Filtered pose of the object in world coordinates. In simulation, this is the same as the unfiltered object_pose.
- Type:
- object_pose
Pose of the object in world coordinates.
- Type: