Lift Pot

Lift Pot#

Grasp a kitchen pot by both handles and lift it above the table without tilting.

Lift Pot Task

Task Description#

The robot must:

  1. Grasp the pot by both handles using both grippers

  2. Lift the pot at least 0.10 m above its initial height

  3. Keep the pot’s symmetry axis within 20° of vertical

  4. Avoid colliding with surrounding cabinets

Key Skills:

  • Bimanual coordination

  • Symmetric grasping

  • Orientation control

  • Spatial awareness

Success Criteria#

  • Pot not colliding with either cabinet

  • Pot lifted ≥ 0.10 m above starting height (SUCCESS_HEIGHT = INITIAL_HEIGHT + 0.1)

  • Pot orientation within 20° of vertical (phi and theta angles checked)

Failure Conditions:

  • Pot colliding with floor

Task Stages#

Stage 0#

Initial state - pot on table, robot ready

Stage 1#

Left gripper grasping the pot handle

Stage 2#

Right gripper grasping the pot handle

Stage 3#

Pot lifted at least 0.10 m above table

Stage 4#

Complete success - pot stable and level

Variations#

LiftPot (Base)#

from roboeval.envs.lift_pot import LiftPot

env = LiftPot(
    action_mode=JointPositionActionMode(),
    robot_cls=BimanualPanda
)

Characteristics:

  • Pot always appears in same position

  • Same orientation every episode

  • Good for initial learning

LiftPotPosition#

from roboeval.envs.lift_pot import LiftPotPosition

Randomization:

  • Position: Random within 0.2m radius

  • Orientation: Fixed

LiftPotOrientation#

from roboeval.envs.lift_pot import LiftPotOrientation

Randomization:

  • Position: Fixed

  • Orientation: ±30° around Z-axis

LiftPotPositionAndOrientation#

from roboeval.envs.lift_pot import LiftPotPositionAndOrientation

Randomization:

  • Position: Random within workspace

  • Orientation: ±30° around Z-axis

  • Most challenging variation

Common Challenges#

Symmetric Grasping

Both grippers must grasp handles simultaneously for balanced lift

Collision Avoidance

Cabinets on both sides require careful arm movements

Orientation Control

Maintaining pot level while lifting requires coordination

Example Usage#

from roboeval.envs.lift_pot import LiftPotPositionAndOrientation
from roboeval.action_modes import JointPositionActionMode
from roboeval.robots.configs.panda import BimanualPanda

env = LiftPotPositionAndOrientation(
    action_mode=JointPositionActionMode(absolute=True),
    render_mode="human",
    robot_cls=BimanualPanda,
    control_frequency=20
)

obs, info = env.reset()

for step in range(1000):
    action = policy.predict(obs)
    obs, reward, terminated, truncated, info = env.step(action)

    if 'metrics' in info:
        print(f"Stage: {info['metrics']['stage']}")

    if terminated or truncated:
        print(f"Success: {info.get('success', False)}")
        break

Metrics#

Standard metrics provided:

  • success: Task completion

  • stage: Current progression stage (0-4)

  • pot_height: Current pot height above table

  • pot_tilt: Pot orientation deviation from vertical

  • collision: Whether collision occurred

Access metrics:

obs, reward, terminated, truncated, info = env.step(action)

if 'metrics' in info:
    metrics = info['metrics']
    print(f"Pot height: {metrics['pot_height']:.3f}m")
    print(f"Pot tilt: {metrics['pot_tilt']:.1f}°")

Demonstrations#

Load demonstrations for this task:

from roboeval.demonstrations.demo_store import DemoStore
from roboeval.demonstrations.utils import Metadata

metadata = Metadata.from_env(env)
demo_store = DemoStore()

# Get demonstrations for base task
demos = demo_store.get_demos(metadata, amount=50)

# Or specify task variation
demos = demo_store.get_demos_for_task("LiftPotPosition", amount=50)

Tips#

  1. Grasp handles symmetrically for better balance

  2. Lift slowly to maintain orientation

  3. Start with base variation before trying randomized versions

  4. Use stage information to debug partial successes

See Also#

Contents