dynibo 0.5.2

Tree-structured robot kinematics and dynamics with runtime-size workspace APIs
Documentation 
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
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208

#!/usr/bin/env python3
"""Franka/G1 public Python API benchmarks against Pinocchio; see README.md."""

from __future__ import annotations

import argparse
from collections import defaultdict, deque
from datetime import datetime, timezone
import gc
import hashlib
import json
import os
from pathlib import Path
import platform
import random
import statistics
import timeit
import xml.etree.ElementTree as ET

import numpy as np
import pinocchio as pin
import dynibo
from dynibo import BaseState, FloatingRobot, Pose, Robot, Twist

ROOT = Path(__file__).resolve().parents[1]
MODELS = (
    ("franka_fixed", "examples/data/franka/franka_fer.urdf", "fer_link8", False, 7),
    ("g1_floating", "examples/data/unitree-g1/g1_29dof_mode_11.urdf", "left_rubber_hand", True, 29),
)


def active_joints(path):
    """Read Dynibo's breadth-first joint order, retaining URDF sibling order.

    Python does not expose joint_name yet. The complete Jacobian/RNEA/ABA
    comparisons below validate this ordering against independently named
    Pinocchio joints; never copy a branched model's vectors without mapping.
    """
    root = ET.parse(path).getroot()
    children = defaultdict(list)
    child_links = set()
    for joint in root.findall("joint"):
        children[joint.find("parent").get("link")].append(joint)
        child_links.add(joint.find("child").get("link"))
    roots = [link.get("name") for link in root.findall("link") if link.get("name") not in child_links]
    assert len(roots) == 1
    queue = deque(roots)
    result = []
    while queue:
        for joint in children[queue.popleft()]:
            queue.append(joint.find("child").get("link"))
            if joint.get("type") != "fixed":
                assert joint.get("type") == "revolute"
                result.append(joint)
    return result


def make_case(spec):
    name, relative, target_name, floating, n = spec
    path = ROOT / relative
    robot = (FloatingRobot if floating else Robot).from_urdf(path)
    joints = active_joints(path)
    assert robot.joint_count == len(joints) == n
    g = robot.generalized_count
    assert g == n + (6 if floating else 0)
    indices = np.arange(1, n + 1, dtype=float)
    lower = np.array([float(j.find("limit").get("lower")) for j in joints])
    upper = np.array([float(j.find("limit").get("upper")) for j in joints])
    q = (lower + upper) * 0.5 + (upper - lower) * 0.1 * np.sin(0.37 * indices)
    v = 0.4 * np.cos(0.23 * indices)
    a = 0.3 * np.sin(0.41 * indices)
    omega, linear = np.array([0.1, -0.05, 0.08]), np.array([0.2, 0.0, -0.1])
    alpha, acc = np.array([0.02, 0.03, -0.01]), np.array([0.1, -0.2, 0.05])
    base = BaseState(Pose(translation=(0.2, -0.1, 0.8)), Twist(omega, linear), Twist(alpha, acc))
    base_args = (base,) if floating else ()
    target = robot.link_id(target_name)
    model = pin.buildModelFromUrdf(str(path), pin.JointModelFreeFlyer()) if floating else pin.buildModelFromUrdf(str(path))
    model.gravity.linear = np.array([0.0, 0.0, -9.80665])
    data = model.createData()
    frame = model.getFrameId(target_name, pin.BODY)
    assert frame < model.nframes and model.nv == g
    pq, pv, pa = pin.neutral(model), np.zeros(g), np.zeros(g)
    columns = [3, 4, 5, 0, 1, 2] if floating else []
    if floating:
        pq[:3] = base.frame.translation
        pv[:3], pv[3:6] = linear, omega
        pa[:3], pa[3:6] = acc - np.cross(omega, linear), alpha
    for i, joint in enumerate(joints):
        jid = model.getJointId(joint.get("name"))
        assert jid < model.njoints
        pj = model.joints[jid]
        assert pj.nq == pj.nv == 1
        pq[pj.idx_q], pv[pj.idx_v], pa[pj.idx_v] = q[i], v[i], a[i]
        columns.append(pj.idx_v)
    forces = robot.inverse_dynamics(*base_args, q, v, a)
    expected = np.r_[alpha, acc, a] if floating else a
    roundtrip = robot.forward_dynamics(*base_args, q, v, forces)
    np.testing.assert_allclose(roundtrip, expected, atol=1e-8, rtol=1e-8)
    forces = forces.copy()
    if floating:
        forces[:6] = 0.0  # Unactuated base, no contact constraints.
    ptau = np.empty(g)
    ptau[columns] = forces
    if floating:
        dyn_functions = {
            "jacobian": lambda: robot.jacobian(base, q, target),
            "rnea": lambda: robot.inverse_dynamics(base, q, v, a),
            "aba": lambda: robot.forward_dynamics(base, q, v, forces),
        }
    else:
        dyn_functions = {
            "jacobian": lambda: robot.jacobian(q, target),
            "rnea": lambda: robot.inverse_dynamics(q, v, a),
            "aba": lambda: robot.forward_dynamics(q, v, forces),
        }
    pin_functions = {
        "jacobian": lambda: pin.computeFrameJacobian(model, data, pq, frame, pin.LOCAL_WORLD_ALIGNED),
        "rnea": lambda: pin.rnea(model, data, pq, pv, pa),
        "aba": lambda: pin.aba(model, data, pq, pv, ptau),
    }
    functions = {op: {"dynibo": dyn_functions[op], "pinocchio": pin_functions[op]}
                 for op in dyn_functions}
    errors = {}
    for operation, functions_pair in functions.items():
        actual = np.array(functions_pair["dynibo"](), copy=True)
        reference = np.array(functions_pair["pinocchio"](), copy=True)
        if operation == "jacobian":
            actual = actual.reshape((6, g), order="F")
            reference = reference[np.ix_([3, 4, 5, 0, 1, 2], columns)]
        else:
            reference = reference[columns]
            if operation == "aba" and floating:
                reference[3:6] += np.cross(omega, linear)
        assert np.isfinite(actual).all() and np.isfinite(reference).all()
        np.testing.assert_allclose(actual, reference, atol=1e-8, rtol=1e-8,
                                   err_msg=f"{name}/{operation}")
        errors[operation] = float(np.max(np.abs(actual - reference)))
    metadata = {"name": name, "path": relative, "sha256": hashlib.sha256(path.read_bytes()).hexdigest(),
                "target": target_name, "joints": n, "generalized_count": g,
                "joint_order": [j.get("name") for j in joints], "q": q.tolist(), "v": v.tolist(), "a": a.tolist(),
                "validation_max_abs_error": errors}
    return functions, metadata


def measure_pair(functions, samples, batch_seconds):
    timers = {name: timeit.Timer(fn) for name, fn in functions.items()}
    iterations = {}
    for name, timer in timers.items():
        number = 1
        while timer.timeit(number) < batch_seconds:
            number *= 2
        iterations[name] = number
    values = {name: [] for name in timers}
    rng = random.Random(20260831)
    for _ in range(samples):
        order = list(timers)
        rng.shuffle(order)
        for name in order:
            ns = timers[name].timeit(iterations[name]) * 1e9 / iterations[name]
            values[name].append(ns)
    return {name: {"median_ns": statistics.median(timings),
                   "p10_ns": float(np.percentile(timings, 10)),
                   "p90_ns": float(np.percentile(timings, 90)),
                   "iterations_per_sample": iterations[name], "samples_ns": timings}
            for name, timings in values.items()}


def main():
    parser = argparse.ArgumentParser(description=__doc__)
    parser.add_argument("--samples", type=int, default=30)
    parser.add_argument("--batch-seconds", type=float, default=0.02)
    parser.add_argument("--cpu", type=int)
    parser.add_argument("--check", action="store_true", help="validate both models without timing")
    parser.add_argument("--output", type=Path, default=ROOT / "target/benchmarks/python.json")
    args = parser.parse_args()
    if args.samples < 3 or args.batch_seconds <= 0:
        parser.error("need at least 3 samples and a positive batch time")
    if args.cpu is not None:
        os.sched_setaffinity(0, {args.cpu})
    native = __import__("dynibo._dynibo", fromlist=["__file__"])
    report = {"timestamp_utc": datetime.now(timezone.utc).isoformat(), "language": "python",
              "python": platform.python_version(), "numpy": np.__version__, "dynibo": dynibo.__version__,
              "dynibo_extension": native.__file__,
              "dynibo_extension_sha256": hashlib.sha256(Path(native.__file__).read_bytes()).hexdigest(),
              "pinocchio": pin.__version__, "platform": platform.platform(),
              "cpu_affinity": sorted(os.sched_getaffinity(0)) if hasattr(os, "sched_getaffinity") else None,
              "output_api": "default", "samples": args.samples,
              "gc_disabled_during_timing": True, "models": [], "results": []}
    for spec in MODELS:
        functions, metadata = make_case(spec)
        report["models"].append(metadata)
        print("validated", spec[0], metadata["validation_max_abs_error"], flush=True)
        if args.check:
            continue
        for operation, pair in functions.items():
            gc.collect()
            stats = measure_pair(pair, args.samples, args.batch_seconds)
            speedup = stats["pinocchio"]["median_ns"] / stats["dynibo"]["median_ns"]
            report["results"].append({"model": spec[0], "operation": operation, **stats, "speedup": speedup})
            print(f"{spec[0]:14} {operation:8} dynibo={stats['dynibo']['median_ns']/1000:.3f} us "
                  f"pinocchio={stats['pinocchio']['median_ns']/1000:.3f} us speedup={speedup:.3f}x", flush=True)
    args.output.parent.mkdir(parents=True, exist_ok=True)
    args.output.write_text(json.dumps(report, indent=2) + "\n")
    print("saved", args.output)


if __name__ == "__main__":
    main()