environments.md

Environments

Enviornments are based on robosuite, and for more information on the details of robosuite, including how to create an environment, please refer to robosuite doc

We describe the tasks included in this repository.

Task Descriptions

We provide a brief description of each environment included in this repo. The current focuse is on manipulation environments.

ToolUse

• Description: A simple tabletop domain to place a hammer into the cabinet and close the cabinet.
• Robot-type: Single-Arm
• Task-type: Single-Task
• Has rewards: Yes
• Has demonstrations: Yes
  • No. of Demonstrations: 100
  • Robot: Panda
  • Data: Link
• Action Space: OSC_POSITION (3) + Gripper (1)
• Observations: Workspace images, Eye-in-Hand images, Proprioception
• Reference: Bottom-Up Skill Discovery from Unsegmented Demonstrations for Long-Horizon Robot Manipulation

HammerPlace

• Description: A simple tabletop domain to place a hammer into the cabinet and close the cabinet.
• Robot-type: Single-Arm
• Task-type: Single-Task
• Has rewards: Yes
• Has demonstrations: Yes
  • No. of Demonstrations: 100
  • Robot: Panda
  • Data: Link
• Action Space: OSC_POSITION (3) + Gripper (1)
• Observations: Workspace images, Eye-in-Hand images, Proprioception
• Reference: Bottom-Up Skill Discovery from Unsegmented Demonstrations for Long-Horizon Robot Manipulation

Kitchen

• Description: A simple tabletop domain to place a hammer into the cabinet and close the cabinet.
• Robot-type: Single-Arm
• Task-type: Single-Task
• Has rewards: Yes
• Has demonstrations: Yes
  • No. of Demonstrations: 100
  • Robot: Panda
  • Data: Link
• Action Space: OSC_POSITION (3) + Gripper (1)
• Observations: Workspace images, Eye-in-Hand images, Proprioception
• Reference: Bottom-Up Skill Discovery from Unsegmented Demonstrations for Long-Horizon Robot Manipulation

MultitaskKitchenDomain

• Description: A simple tabletop domain to place a hammer into the cabinet and close the cabinet.
• Robot-type: Single-Arm
• Task-type: Multi-Task
  • No. of Tasks: 3
• Has rewards: Yes
• Has demonstrations: Yes
  • No. of Demonstrations in total: 360
  • No. of Demonstrations each task: 120
  • Robot: Panda
  • Data: Link
• Action Space: OSC_POSITION (3) + Gripper (1)
• Observations: Workspace images, Eye-in-Hand images, Proprioception
• Reference: Bottom-Up Skill Discovery from Unsegmented Demonstrations for Long-Horizon Robot Manipulation