Manipulation Tasks

Manipulation Tasks#

The manipulation domain combines RL environments, Task Program expert demonstrations, and environment-only tasks. Tasks directly under manipulation/ are listed here; object and tableware tasks are grouped on a separate child page.

Task

Environment ID and deployment

Embodiment

Capability

Hand Over

HandOver-v1

Dual UR5 with PGI-140-80 grippers

Task Program expert demo

Open Drawer

TaskProgramOpenDrawer-v1 (UR5) and TaskProgramOpenDrawer-Franka-v1 (Franka); choose default or newton from the same task config with --physics

UR5 or Franka Panda

Task Program expert demo

Push Cube

PushCubeRL

UR robot

RL

Repeated Pick and Place

TaskProgramRepeatedPickPlace-v1 and TaskProgramRepeatedPickPlace-Franka-v1; choose default or newton from the same task config with --physics

UR5 or Franka Panda

Task Program expert demo

Use show-task before launch when a logical task has several deployments:

embodichain show-task embodichain_tasks:repeated_pick_place
embodichain run-env \
    --gym_config embodichain_tasks/configs/tasks/manipulation/repeated_pick_place/task.franka.yaml --physics newton

The UR5 deployment keeps its augmentation settings in the task-facing expansion/repeated_pick_place.yaml file referenced by expansion.config; the physical default and newton components stay under envs/.