New inflation layer with optional OpenMP acceleration

.circleci/config.yml

@@ -499,6 +499,7 @@ _environments:
     RCUTILS_LOGGING_USE_STDOUT: "0"
     DEBIAN_FRONTEND: "noninteractive"
     PYTHONUNBUFFERED: "1"
+    OMP_NUM_THREADS: "4"
 
 executors:
   release_exec:

nav2_costmap_2d/CMakeLists.txt

@@ -27,6 +27,13 @@ find_package(nav2_ros_common REQUIRED)
 find_package(tf2_sensor_msgs REQUIRED)
 find_package(visualization_msgs REQUIRED)
 find_package(point_cloud_transport REQUIRED)
+find_package(Eigen3 REQUIRED)
+
+option(ENABLE_OPENMP "Enable OpenMP parallelization" OFF)
+
+if(ENABLE_OPENMP)
+  find_package(OpenMP REQUIRED)
+endif()
 
 nav2_package()
 
@@ -66,6 +73,7 @@ target_link_libraries(nav2_costmap_2d_core PUBLIC
 
 add_library(layers SHARED
   plugins/inflation_layer.cpp
+  plugins/legacy_inflation_layer.cpp
   plugins/static_layer.cpp
   plugins/obstacle_layer.cpp
   src/observation_buffer.cpp
@@ -90,7 +98,19 @@ target_link_libraries(layers PUBLIC
   tf2::tf2
   nav2_ros_common::nav2_ros_common
   point_cloud_transport::point_cloud_transport
+  Eigen3::Eigen
 )
+
+# Link OpenMP if enabled
+if(ENABLE_OPENMP)
+  message(STATUS "OpenMP ENABLED for nav2_costmap_2d layers (${OpenMP_CXX_VERSION})")
+  target_link_libraries(layers PRIVATE OpenMP::OpenMP_CXX)
+  target_compile_options(layers PRIVATE -fopenmp)
+else()
+  message(STATUS "OpenMP DISABLED for nav2_costmap_2d layers")
+  target_compile_options(layers PRIVATE -Wno-unknown-pragmas)
+endif()
+
 target_link_libraries(layers PRIVATE
   pluginlib::pluginlib
 )
@@ -226,5 +246,8 @@ ament_export_dependencies(
   nav2_ros_common
   point_cloud_transport
 )
+
+ament_export_dependencies(Eigen3)
+
 pluginlib_export_plugin_description_file(nav2_costmap_2d costmap_plugins.xml)
 ament_package()

nav2_costmap_2d/costmap_plugins.xml

@@ -1,7 +1,10 @@
 <class_libraries>
   <library path="layers">
     <class type="nav2_costmap_2d::InflationLayer"  base_class_type="nav2_costmap_2d::Layer">
-      <description>Inflates obstacles to speed collision checking and to make robot prefer to stay away from obstacles.</description>
+      <description>OpenMP-accelerated inflation layer using Felzenszwalb-Huttenlocher distance transform for improved performance.</description>
+    </class>
+    <class type="nav2_costmap_2d::LegacyInflationLayer"  base_class_type="nav2_costmap_2d::Layer">
+      <description>Legacy inflation layer using traditional CellData-based algorithm (original Navigation2 implementation).</description>
     </class>
     <class type="nav2_costmap_2d::ObstacleLayer"   base_class_type="nav2_costmap_2d::Layer">
       <description>Listens to laser scan and point cloud messages and marks and clears grid cells.</description>

nav2_costmap_2d/include/nav2_costmap_2d/distance_transform.hpp

@@ -0,0 +1,179 @@
+// Copyright (c) 2026, Dexory (Tony Najjar)
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#ifndef NAV2_COSTMAP_2D__DISTANCE_TRANSFORM_HPP_
+#define NAV2_COSTMAP_2D__DISTANCE_TRANSFORM_HPP_
+
+#include <limits>
+#include <vector>
+#ifdef _OPENMP
+#include <omp.h>
+#endif
+#include <Eigen/Core>
+
+
+namespace nav2_costmap_2d
+{
+
+/// Row-major float matrix type for efficient row-wise access
+using MatrixXfRM = Eigen::Matrix<float, Eigen::Dynamic, Eigen::Dynamic, Eigen::RowMajor>;
+
+/**
+ * @class DistanceTransform
+ * @brief Efficient Euclidean distance transform using the Felzenszwalb-Huttenlocher algorithm.
+ *
+ * This class provides a standalone implementation of the linear-time (O(n)) distance transform
+ * algorithm based on the lower envelope of parabolas method. It can be used for various
+ * applications including costmap inflation, obstacle detection, and path planning.
+ *
+ * The algorithm computes exact squared Euclidean distances in two separable 1D passes
+ * (rows and columns), making it highly efficient for large images/grids.
+ *
+ * Reference: Distance Transforms of Sampled Functions
+ * P. Felzenszwalb and D. Huttenlocher
+ * Theory of Computing, Vol. 8, No. 19, September 2012
+ */
+class DistanceTransform
+{
+public:
+  /// Infinity constant for distance transform
+  static constexpr float DT_INF = std::numeric_limits<float>::max();
+
+  /**
+   * @brief Perform 1D distance transform using lower envelope of parabolas.
+   *
+   * This is the core Felzenszwalb-Huttenlocher algorithm for computing squared
+   * Euclidean distances in 1D. It uses a lower envelope of parabolas to achieve
+   * linear time complexity.
+   *
+   * @param f Input array of squared distances (typically 0 for obstacles, INF for free space)
+   * @param d Output array for transformed squared distances (same size as f)
+   * @param n Length of the arrays
+   * @param v Buffer for parabola indices (size n)
+   * @param z Buffer for parabola boundaries (size n+1)
+   */
+  static void distanceTransform1D(
+    const float * f, float * d, int n,
+    int * v, float * z)
+  {
+    if (!f || !d || !v || !z || n <= 0) {
+      return;
+    }
+
+    int k = 0;
+    v[0] = 0;
+    z[0] = -DT_INF;
+    z[1] = DT_INF;
+
+    for (int q = 1; q < n; q++) {
+      // Use integer arithmetic for squared values to avoid precision loss
+      // Only convert to float for the division operation
+      float s = (f[q] - f[v[k]] + static_cast<float>(q * q - v[k] * v[k])) /
+        (2.0f * static_cast<float>(q - v[k]));
+      while (s <= z[k]) {
+        k--;
+        s = (f[q] - f[v[k]] + static_cast<float>(q * q - v[k] * v[k])) /
+          (2.0f * static_cast<float>(q - v[k]));
+      }
+      k++;
+      v[k] = q;
+      z[k] = s;
+      z[k + 1] = DT_INF;
+    }
+
+    k = 0;
+    for (int q = 0; q < n; q++) {
+      while (z[k + 1] < static_cast<float>(q)) {
+        k++;
+      }
+      const int diff = q - v[k];
+      d[q] = static_cast<float>(diff * diff) + f[v[k]];
+    }
+  }
+
+  /**
+   * @brief Perform 2D Euclidean distance transform using separable passes.
+   *
+   * This method applies the 1D distance transform separately along columns and rows,
+   * exploiting the separability property of the Euclidean metric. The result is an
+   * exact Euclidean distance map computed in O(width × height) time.
+   *
+   * The algorithm is parallelized using OpenMP when available for improved performance
+   * on multi-core systems.
+   *
+   * @param img Input/output matrix (modified in place). Input values should be 0 for
+   *            obstacles and DT_INF for free space. Output will contain Euclidean distances.
+   * @param height Number of rows in the matrix
+   * @param width Number of columns in the matrix
+   */
+  static void distanceTransform2D(MatrixXfRM & img, int height, int width)
+  {
+    // Column pass (parallelizable)
+#ifdef _OPENMP
+    #pragma omp parallel for schedule(dynamic, 16)
+#endif
+    for (int x = 0; x < width; x++) {
+      // Thread-local buffers
+      std::vector<float> f(height);
+      std::vector<float> d(height);
+      std::vector<int> v(height);
+      std::vector<float> z(height + 1);
+
+      // Extract column
+      for (int y = 0; y < height; y++) {
+        f[y] = img(y, x);
+      }
+
+      // 1D transform
+      distanceTransform1D(f.data(), d.data(), height, v.data(), z.data());
+
+      // Write back
+      for (int y = 0; y < height; y++) {
+        img(y, x) = d[y];
+      }
+    }
+
+    // Row pass (parallelizable)
+#ifdef _OPENMP
+    #pragma omp parallel for schedule(dynamic, 16)
+#endif
+    for (int y = 0; y < height; y++) {
+      // Thread-local buffers
+      std::vector<float> f(width);
+      std::vector<float> d(width);
+      std::vector<int> v(width);
+      std::vector<float> z(width + 1);
+
+      // Extract row (already contiguous in row-major)
+      for (int x = 0; x < width; x++) {
+        f[x] = img(y, x);
+      }
+
+      // 1D transform
+      distanceTransform1D(f.data(), d.data(), width, v.data(), z.data());
+
+      // Write back
+      for (int x = 0; x < width; x++) {
+        img(y, x) = d[x];
+      }
+    }
+
+    // Square root to get Euclidean distance (not squared distance)
+    img = img.cwiseSqrt();
+  }
+};
+
+}  // namespace nav2_costmap_2d
+
+#endif  // NAV2_COSTMAP_2D__DISTANCE_TRANSFORM_HPP_