Adroit Door#
+Adroit Door¶
+
+
Description#
+Description¶
This environment was introduced in “Learning Complex Dexterous Manipulation with Deep Reinforcement Learning and Demonstrations” by Aravind Rajeswaran, Vikash Kumar, Abhishek Gupta, Giulia Vezzani, John Schulman, Emanuel Todorov, and Sergey Levine.
The environment is based on the Adroit manipulation platform, a 28 degree of freedom system which consists of a 24 degrees of freedom @@ -331,7 +333,7 @@
Description
-Action Space#
+Action Space¶
The action space is a Box(-1.0, 1.0, (28,), float32). The control actions are absolute angular positions of the Adroit hand joints. The input of the control actions is set to a range between -1 and 1 by scaling the real actuator angle ranges in radians.
The elements of the action array are the following:
@@ -634,7 +636,7 @@ Action Space
-Observation Space#
+Observation Space¶
The observation space is of the type Box(-inf, inf, (39,), float64). It contains information about the angular position of the finger joints, the pose of the palm of the hand, as well as state of the latch and door.
Box(-1.0, 1.0, (28,), float32). The control actions are absolute angular positions of the Adroit hand joints. The input of the control actions is set to a range between -1 and 1 by scaling the real actuator angle ranges in radians.
The elements of the action array are the following:Action Space
-Observation Space#
+Observation Space¶
The observation space is of the type Box(-inf, inf, (39,), float64). It contains information about the angular position of the finger joints, the pose of the palm of the hand, as well as state of the latch and door.
Box(-inf, inf, (39,), float64). It contains information about the angular position of the finger joints, the pose of the palm of the hand, as well as state of the latch and door.