import mujoco
import mujoco_warp as mjw
import warp as wp
wp.config.quiet = True
import numpy as np
np.set_printoptions(precision=4, suppress=True)
scene = """
<mujoco model="MuJoCo Model">
<option gravity="1.0 0 -1.0" timestep="0.01" integrator="RK4" iterations="100" tolerance="0" >
</option>
<worldbody>
<geom name="ground" type="plane" size="2 1 1" pos="0 0 0" quat="1 0 0 0" condim="3" friction="0 0 0"/>
<body name="box_1" pos="0.0 0.0 0.005">
<freejoint />
<inertial pos="0 0 0" mass="1" diaginertia="0.0008416666666666668 0.0008416666666666668 0.001666666666666667"/>
<geom name="box_1_geom_0" size="0.05 0.05 0.005" type="box" condim="3" friction="0 0 0"/>
</body>
</worldbody>
</mujoco>
"""
nsteps = 10 # simulate in 10 steps
# sim in mujoco
mj_model = mujoco.MjModel.from_xml_string(scene)
mj_data = mujoco.MjData(mj_model)
for _ in range(nsteps):
mujoco.mj_step(mj_model, mj_data)
mj_qpos = mj_data.qpos.copy()
# sim in mujoco warp
mj_model = mujoco.MjModel.from_xml_string(scene)
mj_data = mujoco.MjData(mj_model)
mjw_model = mjw.put_model(mj_model)
mjw_data = mjw.put_data(mj_model, mj_data, njmax=100, nworld=1)
with wp.ScopedCapture() as capture:
mjw.step(mjw_model, mjw_data)
for _ in range(nsteps):
wp.capture_launch(capture.graph)
mjw_qpos = mjw_data.qpos.numpy().copy()
print("in mujoco, qpos is:\n", mj_qpos)
print("in mujoco warp, qpos is:\n", mjw_qpos)output is nan value
in mujoco, qpos is:
[0.005 0. 0.005 1. 0. 0. 0. ]
in mujoco warp, qpos is:
[[nan nan nan 0. 0. 0. 1.]]
output is nan value