Grasp Planning with a RealHand L6¶
This walkthrough exports a six-coordinate right RealHand L6 and
plans grasps on a 50 mm cube. It requires the image from
Installation and a local right-hand MuJoCo scene with
collision meshes. The reference hand uses palm body
l6_hand_base_link and the following independent joint order.
l6_thumb_cmc_roll
l6_thumb_cmc_pitch
l6_index_mcp_pitch
l6_middle_mcp_pitch
l6_ring_mcp_pitch
l6_pinky_mcp_pitch
Export the Hand and Cube¶
Place scene.xml and the associated meshes under input/ and
preserve the relative mesh paths. Check the palm and thumb joint
names against the source scene. Run the export from the repository
root.
mkdir -p output
docker run --rm -i --user "$(id -u):$(id -g)" \
-v "$PWD/input:/input:ro" -v "$PWD/output:/output" graspit-py python - <<'PYTHON'
import mujoco
from graspit_python_wrapper.convert.mjcf_hand import export_body, export_hand
model = mujoco.MjModel.from_xml_path("/input/scene.xml")
hand = export_hand(model, "l6_hand_base_link", "/output/hand", name="l6",
preshape_joints=["l6_thumb_cmc_roll"])
cube = mujoco.MjModel.from_xml_string("""
<mujoco>
<worldbody>
<body name="cube">
<geom type="box" size="0.025 0.025 0.025" mass="0.100"/>
</body>
</worldbody>
</mujoco>
""")
export_body(cube, "cube", "/output/object")
print(hand.dof_joints)
PYTHON
Generate and Inspect Grasps¶
Compare the printed DOF order with the order above before applying the included preshapes. The configuration places the support plane 25 mm below the cube center and specifies the approach axes and grasp center in the reference palm frame. For another model, adapt these three frame parameters and express each preshape in its exported DOF order.
docker run --rm --user "$(id -u):$(id -g)" \
-v "$PWD/output:/output" graspit-py \
xvfb-run -a python -m graspit_python_wrapper.planning \
--robot /output/hand/l6.xml --object /output/object/cube.xml \
--config /src/docs/examples/realhand-l6-cube.json \
--top-k 5 --out /output/candidates.json
{
"hand": {
"approach_axis": [
1,
0,
0
],
"finger_axis": [
0,
0,
1
],
"grasp_center_mm": [
40.1,
10.3,
93.1
]
},
"planner": {
"n_directions": 40,
"n_rolls": 6,
"standoff_mm": 180.0,
"tail_mass": 0.5,
"friction_samples": [
0.4,
0.7,
1.0
],
"seed": 42,
"n_pose_samples": 4,
"support_normal": [
0,
0,
1
],
"support_offset_mm": -25.0,
"preshapes": [
[
0.5847,
0.0,
0.0,
0.0,
0.0,
0.0
],
[
0.5847,
0.113088,
0.16336799999999999,
0.16336799999999999,
0.16336799999999999,
0.16336799999999999
],
[
0.8535,
0.0,
0.0,
0.0,
0.0,
0.0
],
[
0.8535,
0.113088,
0.16336799999999999,
0.16336799999999999,
0.16336799999999999,
0.16336799999999999
],
[
1.1222,
0.0,
0.0,
0.0,
0.0,
0.0
],
[
1.1222,
0.113088,
0.16336799999999999,
0.16336799999999999,
0.16336799999999999,
0.16336799999999999
]
]
}
}
The CLI writes ranked grasps to output/candidates.json. The first
record has the greatest ranking score among the admitted grasps.
Inspect its contact count and score using Python’s standard library.
For an empty list, the top-level rejections field identifies the
geometric criteria the nominal approaches failed to meet.
import json
from pathlib import Path
result = json.loads(Path("output/candidates.json").read_text())
if result["candidates"]:
candidate = result["candidates"][0]
print(len(candidate["contacts"]), candidate["epsilon_cvar"])
else:
print(result["rejections"])
See grasp planning for score and pose conventions and MuJoCo export for reconstructing a selected grasp in the source model.