Lift

Task key: lift.

lift is a single-arm baseline task. A cube is spawned inside the robot workspace and the agent must lift it to a fixed target height.

Goal

• Lift the cube until it reaches the goal height.

• Keep the cube’s XY position close to the goal marker.

Quickstart

import dexsuite as ds

env = ds.make(
    "lift",
    manipulator="franka",
    gripper="robotiq",
    arm_control="osc_pose",
    gripper_control="joint_position",
    render_mode="human",
)

obs, info = env.reset()
obs, reward, terminated, truncated, info = env.step(env.action_space.sample())

env.close()

Task Entities

• cube: a rigid cube that starts on the table.

• goal: a fixed, non-colliding sphere placed above the cube spawn.

Observations

Observations follow the DexSuite convention:

• obs["state"] is a nested dict of torch tensors (robot state + task state).

• obs["cameras"] is a dict of rendered camera outputs (if cameras are enabled).

Lift adds (under obs["state"]["other"]):

• cube_pos (shape (n_envs, 3))

• cube_quat (shape (n_envs, 4), wxyz convention)

• target_pos (shape (n_envs, 3))

The base environment also adds:

• other["action"] and other["last_action"] (flat action tensors)

Actions

DexSuite uses a single, flat action tensor:

• Single env: shape (D,)

• Batched envs: shape (n_envs, D)

D depends on your robot + controller choices. Use env.action_space.shape to inspect it.

env.step(...) accepts a torch tensor, a NumPy array, or a Python sequence. Inputs are converted to a float32 torch tensor on the environment device.

Termination

• Success: cube reaches the target height (within tolerance) and its XY position is within 5 cm of the target.

• Failure: the tool center point (TCP) leaves the workspace AABB.

• Truncation: the episode reaches the horizon.

Simulation Settings

The lift baseline uses:

• SIM_DT = 0.01

• SUBSTEPS = 1

• HORIZON = 600 control steps

SimOptions.control_hz controls the control frequency; physics steps run at SIM_DT with SUBSTEPS substeps per physics step.

Source

• Dexsuite/dexsuite/environments/single/baselines/lift.py