mink/examples/kinetic_energy_reg.py at main · jaikr-dev/mink · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
"""UR5e figure-8 path with kinetic energy regularization.

Example usage:

    python examples/kinetic_energy_reg.py --help
    python examples/kinetic_energy_reg.py --energy_reg 0.0   # No regularization.
    python examples/kinetic_energy_reg.py --energy_reg 1e-5  # Low regularization.
"""

import argparse
from collections import deque
from pathlib import Path
from typing import Deque

import mujoco
import mujoco.viewer
import numpy as np
from loop_rate_limiters import RateLimiter

import mink

_HERE = Path(__file__).parent
_XML = _HERE / "universal_robots_ur5e" / "scene_plain.xml"

# IK integration timestep, in [s].
_DT = 0.02

# Maximum number of trace points to plot.
_DESIRED_TRACE_DURATION = 5.0  # [s]
_MAX_TRACE_POINTS = int(_DESIRED_TRACE_DURATION / _DT)

# Trace visualization parameters.
_RGBA = np.array([0, 1, 0.5, 0.8])
_RADIUS = 0.003


if __name__ == "__main__":
    parser = argparse.ArgumentParser(
        description="UR5e figure-8 path with kinetic energy regularization."
    )
    parser.add_argument(
        "--energy_reg",
        type=float,
        default=0.0,
        help="Regularization weight for the kinetic energy task.",
    )
    args = parser.parse_args()

    model = mujoco.MjModel.from_xml_path(_XML.as_posix())
    configuration = mink.Configuration(model)

    end_effector_task = mink.FrameTask(
        frame_name="attachment_site",
        frame_type="site",
        position_cost=1.0,
        orientation_cost=0.0,
    )

    kinetic_energy_task = mink.KineticEnergyRegularizationTask(cost=args.energy_reg)
    kinetic_energy_task.set_dt(_DT)  # NOTE: This is required!

    # For storing and visualizing the end-effector path.
    positions: Deque[np.ndarray] = deque(maxlen=_MAX_TRACE_POINTS)

    def add_visual_capsule(scene, point1, point2, radius, rgba):
        if scene.ngeom >= scene.maxgeom:
            return
        scene.ngeom += 1
        mujoco.mjv_initGeom(
            scene.geoms[scene.ngeom - 1],
            mujoco.mjtGeom.mjGEOM_CAPSULE,
            np.zeros(3),
            np.zeros(3),
            np.zeros(9),
            rgba.astype(np.float32),
        )
        mujoco.mjv_connector(
            scene.geoms[scene.ngeom - 1],
            mujoco.mjtGeom.mjGEOM_CAPSULE,
            radius,
            point1,
            point2,
        )

    def modify_scene(scn):
        scn.ngeom = 0
        for i in range(len(positions) - 1):
            if np.allclose(positions[i], positions[i + 1]):
                continue
            add_visual_capsule(scn, positions[i], positions[i + 1], _RADIUS, _RGBA)

    solver = "daqp"
    model = configuration.model
    data = configuration.data

    # Do an initial solve to find the initial configuration that achieves the target
    # position.
    z = 0.25
    pos0 = np.array([0.5, 0.0, z])
    configuration.update_from_keyframe("home")
    end_effector_task.set_target(mink.SE3.from_translation(pos0))
    for _ in range(10):
        vel = mink.solve_ik(configuration, [end_effector_task], _DT, solver)
        configuration.integrate_inplace(vel, _DT)
    qpos0 = configuration.q.copy()

    tasks = [end_effector_task, kinetic_energy_task]

    with mujoco.viewer.launch_passive(
        model=model, data=data, show_left_ui=False, show_right_ui=False
    ) as viewer:
        mujoco.mjv_defaultFreeCamera(model, viewer.cam)

        configuration.update(qpos0)
        mink.move_mocap_to_frame(model, data, "target", "attachment_site", "site")

        rate = RateLimiter(frequency=(1.0 / _DT), warn=False)
        t = 0.0
        while viewer.is_running():
            # Update task target using a figure-8 pattern.
            x = 0.5 + 0.1 * np.sin(2 * t)
            y = 0.2 * np.sin(t)
            data.mocap_pos[0] = np.array([x, y, z])
            T_wt = mink.SE3.from_mocap_name(model, data, "target")
            end_effector_task.set_target(T_wt)

            # Compute velocity and integrate into the next configuration.
            vel = mink.solve_ik(configuration, tasks, _DT, solver)
            configuration.integrate_inplace(vel, _DT)
            mujoco.mj_camlight(model, data)
            positions.append(data.site_xpos[data.site("attachment_site").id].copy())

            # Visualize at fixed FPS.
            modify_scene(viewer.user_scn)
            viewer.sync()
            rate.sleep()
            t += _DT