Merge branch 'devel' into topic/coal

Author: jcarpent

.github/workflows/update-flake-lock.yml

@@ -0,0 +1,17 @@
+name: update-flake-lock
+
+on:
+  workflow_dispatch:
+  schedule:
+    - cron: '0 1 5 * *'
+
+jobs:
+  lockfile:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+      - name: Install Nix
+        uses: DeterminateSystems/nix-installer-action@main
+      - name: Update flake.lock
+        uses: DeterminateSystems/update-flake-lock@main

.pre-commit-config.yaml

@@ -4,7 +4,7 @@ ci:
   autoupdate_schedule: quarterly
 repos:
   - repo: https://github.com/pre-commit/mirrors-clang-format
-    rev: v19.1.1
+    rev: v19.1.6
     hooks:
       - id: clang-format
         types_or: []
@@ -27,7 +27,7 @@ repos:
               doc/doxygen-awesome.*
           )$
   - repo: https://github.com/astral-sh/ruff-pre-commit
-    rev: v0.6.9
+    rev: v0.8.6
     hooks:
       - id: ruff
       - id: ruff-format

CHANGELOG.md

@@ -9,6 +9,12 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/).
 ### Fixed
 - Fix mjcf Euler angle parsing: use xyz as a default value for eulerseq compiler option ([#2526](https://github.com/stack-of-tasks/pinocchio/pull/2526))
 - Fix variable naming in Python ([#2530](https://github.com/stack-of-tasks/pinocchio/pull/2530))
+- Fix aba explicit template instantiation ([#2541](https://github.com/stack-of-tasks/pinocchio/pull/2541))
+- Add parsing meshes with vertices for MJCF format ([#2537](https://github.com/stack-of-tasks/pinocchio/pull/2537))
+- CMake: fix RPATH on macos ([#2546](https://github.com/stack-of-tasks/pinocchio/pull/2546))
+- Fix aba explicit template instantiation ([#2541](https://github.com/stack-of-tasks/pinocchio/pull/2541))
+- Fix mjcf parsing of keyframe qpos with newlines ([#2535](https://github.com/stack-of-tasks/pinocchio/pull/2535))
+
 
 ## [3.3.1] - 2024-12-13
 

bindings/python/CMakeLists.txt

@@ -115,7 +115,7 @@ function(PINOCCHIO_PYTHON_BINDINGS_SPECIFIC_TYPE scalar_name)
                # On Windows, shared library are treat as binary
                RUNTIME_OUTPUT_DIRECTORY "${PROJECT_BINARY_DIR}/bindings/python/${PROJECT_NAME}")
 
-  if(UNIX AND NOT APPLE)
+  if(UNIX)
     get_relative_rpath(${${PYWRAP}_INSTALL_DIR} ${PYWRAP}_INSTALL_RPATH)
     set_target_properties(${PYTHON_LIB_NAME} PROPERTIES INSTALL_RPATH "${${PYWRAP}_INSTALL_RPATH}")
   endif()

bindings/python/pinocchio/shortcuts.py

@@ -5,7 +5,8 @@
 
 ## In this file, some shortcuts are provided ##
 
-from typing import Tuple
+# TODO: Remove when 20.04 is not supported
+from __future__ import annotations
 
 from . import WITH_HPP_FCL, WITH_HPP_FCL_BINDINGS
 from . import pinocchio_pywrap_default as pin
@@ -15,7 +16,7 @@
 
 def buildModelsFromUrdf(
     filename, *args, **kwargs
-) -> Tuple[pin.Model, pin.GeometryModel, pin.GeometryModel]:
+) -> tuple[pin.Model, pin.GeometryModel, pin.GeometryModel]:
     """Parse the URDF file given in input and return a Pinocchio Model followed by corresponding GeometryModels of types specified by geometry_types, in the same order as listed.
     Arguments:
         - filename - name of the urdf file to load
@@ -63,7 +64,7 @@ def _buildModelsFromUrdf(
     verbose=False,
     meshLoader=None,
     geometry_types=None,
-) -> Tuple[pin.Model, pin.GeometryModel, pin.GeometryModel]:
+) -> tuple[pin.Model, pin.GeometryModel, pin.GeometryModel]:
     if geometry_types is None:
         geometry_types = [pin.GeometryType.COLLISION, pin.GeometryType.VISUAL]
 
@@ -119,7 +120,7 @@ def createDatas(*models):
 
 def buildModelsFromSdf(
     filename, *args, **kwargs
-) -> Tuple[pin.Model, pin.GeometryModel, pin.GeometryModel]:
+) -> tuple[pin.Model, pin.GeometryModel, pin.GeometryModel]:
     """Parse the Sdf file given in input and return a Pinocchio Model and a list of Constraint Models, followed by corresponding GeometryModels of types specified by geometry_types, in the same order as listed.
     Arguments:
         - filename - name of the urdf file to load

bindings/python/pinocchio/utils.py

@@ -96,17 +96,17 @@ def fromListToVectorOfString(items):
 
 
 __all__ = [
-    "np",
-    "npl",
     "eye",
-    "zero",
-    "rand",
+    "fromListToVectorOfString",
     "isapprox",
+    "matrixToRpy",
     "mprint",
+    "np",
     "npToTTuple",
     "npToTuple",
+    "npl",
+    "rand",
     "rotate",
     "rpyToMatrix",
-    "matrixToRpy",
-    "fromListToVectorOfString",
+    "zero",
 ]

bindings/python/pinocchio/visualize/meshcat_visualizer.py

@@ -1,6 +1,9 @@
+# TODO: Remove when 20.04 is not supported
+from __future__ import annotations
+
 import warnings
 from pathlib import Path
-from typing import ClassVar, List
+from typing import ClassVar
 
 import numpy as np
 
@@ -21,9 +24,10 @@
 
 # DaeMeshGeometry
 import xml.etree.ElementTree as Et
-from typing import Any, Dict, Optional, Set, Union
+from typing import Any
 
-MsgType = Dict[str, Union[str, bytes, bool, float, "MsgType"]]
+# TODO: Remove quote when 20.04 is not supported
+MsgType = "dict[str, Union[str, bytes, bool, float, 'MsgType']]"
 
 try:
     import hppfcl
@@ -110,7 +114,7 @@ def lower(self, object_data: Any) -> MsgType:
             }
 
     class DaeMeshGeometry(mg.ReferenceSceneElement):
-        def __init__(self, dae_path: str, cache: Optional[Set[str]] = None) -> None:
+        def __init__(self, dae_path: str, cache: set[str] | None = None) -> None:
             """Load Collada files with texture images.
             Inspired from
             https://gist.github.com/danzimmerman/a392f8eadcf1166eb5bd80e3922dbdc5
@@ -131,7 +135,7 @@ def __init__(self, dae_path: str, cache: Optional[Set[str]] = None) -> None:
                 self.dae_raw = text_file.read()
 
             # Parse the image resource in Collada file
-            img_resource_paths: List[Path] = []
+            img_resource_paths: list[Path] = []
             img_lib_element = Et.parse(dae_path).find(
                 "{http://www.collada.org/2005/11/COLLADASchema}library_images"
             )
@@ -143,7 +147,7 @@ def __init__(self, dae_path: str, cache: Optional[Set[str]] = None) -> None:
                 ]
 
             # Convert textures to data URL for Three.js ColladaLoader to load them
-            self.img_resources: Dict[str, str] = {}
+            self.img_resources: dict[str, str] = {}
             for img_path in img_resource_paths:
                 img_key = str(img_path)
                 # Return empty string if already in cache
@@ -164,7 +168,7 @@ def __init__(self, dae_path: str, cache: Optional[Set[str]] = None) -> None:
                 img_uri = f"data:image/png;base64,{img_data.decode('utf-8')}"
                 self.img_resources[img_key] = img_uri
 
-        def lower(self) -> Dict[str, Any]:
+        def lower(self) -> dict[str, Any]:
             """Pack data into a dictionary of the format that must be passed to
             `Visualizer.window.send`.
             """
@@ -1112,10 +1116,10 @@ def drawFrameVelocities(self, frame_id: int, v_scale=0.2, color=FRAME_VEL_COLOR)
 
     def _draw_vectors_from_frame(
         self,
-        vecs: List[np.ndarray],
-        frame_ids: List[int],
-        vec_names: List[str],
-        colors: List[int],
+        vecs: list[np.ndarray],
+        frame_ids: list[int],
+        vec_names: list[str],
+        colors: list[int],
     ):
         """Draw vectors extending from given frames."""
         import meshcat.geometry as mg

development/scripts/misc/common_symbols.py

@@ -5,10 +5,9 @@
 import itertools
 import pathlib
 import subprocess
-import typing
 
 
-def generate_symbols(shared_library: pathlib.Path) -> typing.Set[str]:
+def generate_symbols(shared_library: pathlib.Path) -> set[str]:
     # Show symbol
     # -D: Dynamic
     # -C: Demangled

doc/d-practical-exercises/src/continuous.py

@@ -15,16 +15,16 @@
 import tflearn
 from pendulum import Pendulum
 
-### --- Random seed
+# --- Random seed
 RANDOM_SEED = int((time.time() % 10) * 1000)
-print("Seed = %d" % RANDOM_SEED)
+print(f"Seed = {RANDOM_SEED}")
 np.random.seed(RANDOM_SEED)
 tf.set_random_seed(RANDOM_SEED)
 random.seed(RANDOM_SEED)
 n_init = tflearn.initializations.truncated_normal(seed=RANDOM_SEED)
 u_init = tflearn.initializations.uniform(minval=-0.003, maxval=0.003, seed=RANDOM_SEED)
 
-### --- Hyper paramaters
+# --- Hyper paramaters
 NEPISODES = 100  # Max training steps
 NSTEPS = 100  # Max episode length
 QVALUE_LEARNING_RATE = 0.001  # Base learning rate for the Q-value Network
@@ -35,13 +35,13 @@
 BATCH_SIZE = 64  # Number of points to be fed in stochastic gradient
 NH1 = NH2 = 250  # Hidden layer size
 
-### --- Environment
+# --- Environment
 env = Pendulum(1)  # Continuous pendulum
 env.withSinCos = True  # State is dim-3: (cosq,sinq,qdot) ...
 NX = env.nobs  # ... training converges with q,qdot with 2x more neurones.
 NU = env.nu  # Control is dim-1: joint torque
 
-### --- Q-value and policy networks
+# --- Q-value and policy networks
 
 
 class QValueNetwork:
@@ -63,7 +63,8 @@ def __init__(self):
         self.x = x  # Network state   <x> input in Q(x,u)
         self.u = u  # Network control <u> input in Q(x,u)
         self.qvalue = qvalue  # Network output  <Q>
-        self.variables = tf.trainable_variables()[nvars:]  # Variables to be trained
+        # Variables to be trained
+        self.variables = tf.trainable_variables()[nvars:]
         self.hidens = [netx1, netx2, netu1, netu2]  # Hidden layers for debug
 
     def setupOptim(self):
@@ -75,7 +76,8 @@ def setupOptim(self):
         self.qref = qref  # Reference Q-values
         self.optim = optim  # Optimizer
         self.gradient = (
-            gradient  # Gradient of Q wrt the control  dQ/du (for policy training)
+            # Gradient of Q wrt the control  dQ/du (for policy training)
+            gradient
         )
         return self
 
@@ -101,7 +103,8 @@ def __init__(self):
 
         self.x = x  # Network input <x> in Pi(x)
         self.policy = policy  # Network output <Pi>
-        self.variables = tf.trainable_variables()[nvars:]  # Variables to be trained
+        # Variables to be trained
+        self.variables = tf.trainable_variables()[nvars:]
 
     def setupOptim(self):
         qgradient = tf.placeholder(tf.float32, [None, NU])
@@ -110,7 +113,8 @@ def setupOptim(self):
             zip(grad, self.variables)
         )
 
-        self.qgradient = qgradient  # Q-value gradient wrt control (input value)
+        # Q-value gradient wrt control (input value)
+        self.qgradient = qgradient
         self.optim = optim  # Optimizer
         return self
 
@@ -122,7 +126,7 @@ def setupTargetAssign(self, nominalNet, tau=UPDATE_RATE):
         return self
 
 
-### --- Replay memory
+# --- Replay memory
 class ReplayItem:
     def __init__(self, x, u, r, d, x2):
         self.x = x
@@ -134,7 +138,7 @@ def __init__(self, x, u, r, d, x2):
 
 replayDeque = deque()
 
-### --- Tensor flow initialization
+# --- Tensor flow initialization
 
 policy = PolicyNetwork().setupOptim()
 policyTarget = PolicyNetwork().setupTargetAssign(policy)
@@ -167,24 +171,26 @@ def rendertrial(maxiter=NSTEPS, verbose=True):
     signal.SIGTSTP, lambda x, y: rendertrial()
 )  # Roll-out when CTRL-Z is pressed
 
-### History of search
+# History of search
 h_rwd = []
 h_qva = []
 h_ste = []
 
-### --- Training
+# --- Training
 for episode in range(1, NEPISODES):
     x = env.reset().T
     rsum = 0.0
 
     for step in range(NSTEPS):
-        u = sess.run(policy.policy, feed_dict={policy.x: x})  # Greedy policy ...
+        # Greedy policy ...
+        u = sess.run(policy.policy, feed_dict={policy.x: x})
         u += 1.0 / (1.0 + episode + step)  # ... with noise
         x2, r = env.step(u)
         x2 = x2.T
         done = False  # pendulum scenario is endless.
 
-        replayDeque.append(ReplayItem(x, u, r, done, x2))  # Feed replay memory ...
+        # Feed replay memory ...
+        replayDeque.append(ReplayItem(x, u, r, done, x2))
         if len(replayDeque) > REPLAY_SIZE:
             replayDeque.popleft()  # ... with FIFO forgetting.
 
@@ -260,7 +266,7 @@ def rendertrial(maxiter=NSTEPS, verbose=True):
 
 # \\\END_FOR episode in range(NEPISODES)
 
-print("Average reward during trials: %.3f" % (sum(h_rwd) / NEPISODES))
+print(f"Average reward during trials: {sum(h_rwd) / NEPISODES:.3f}")
 rendertrial()
 plt.plot(np.cumsum(h_rwd) / range(1, NEPISODES))
 plt.show()

doc/d-practical-exercises/src/ocp.py

@@ -36,7 +36,7 @@ def display(U, verbose=False):
         env.display(x)
         time.sleep(5e-2)
         if verbose:
-            print("X%d" % i, x.T)
+            print(f"X{i}")
 
 
 class CallBack:
@@ -66,8 +66,9 @@ def setWithDisplay(self, boolean=None):
 callback = CallBack()
 signal.signal(signal.SIGTSTP, lambda x, y: callback.setWithDisplay())
 
-### --- OCP resolution
-U0 = np.zeros(NSTEPS * env.nu) - env.umax  # Initial guess for the control trajectory.
+# --- OCP resolution
+# Initial guess for the control trajectory.
+U0 = np.zeros(NSTEPS * env.nu) - env.umax
 bounds = (
     [
         [-env.umax, env.umax],