feat(autoware_cuda_pointcloud_preprocessor): a cuda-accelerated pointcloud preprocessor

sensing/autoware_cuda_pointcloud_preprocessor/CMakeLists.txt

@@ -0,0 +1,100 @@
+cmake_minimum_required(VERSION 3.14)
+project(autoware_cuda_pointcloud_preprocessor)
+
+find_package(ament_cmake_auto REQUIRED)
+find_package(CUDA)
+find_package(PCL REQUIRED)
+
+if(NOT ${CUDA_FOUND})
+  message(WARNING "cuda was not found, so the autoware_cuda_pointcloud_preprocessor package will not be built.")
+  return()
+endif()
+
+ament_auto_find_build_dependencies()
+
+# Default to C++17
+if (NOT CMAKE_CXX_STANDARD)
+    set(CMAKE_CXX_STANDARD 17)
+endif ()
+
+if (CMAKE_COMPILER_IS_GNUCXX OR CMAKE_CXX_COMPILER_ID MATCHES "Clang")
+    add_compile_options(-Wall -Wextra -Wpedantic -Wunused-function)
+endif ()
+
+if(BUILD_TESTING)
+    find_package(ament_lint_auto REQUIRED)
+    ament_lint_auto_find_test_dependencies()
+endif()
+
+include_directories(
+    include
+    SYSTEM
+    ${CUDA_INCLUDE_DIRS}
+)
+
+list(APPEND CUDA_NVCC_FLAGS "--expt-relaxed-constexpr -diag-suppress 20012") # cSpell: ignore expt
+
+cuda_add_library(cuda_pointcloud_preprocessor_lib SHARED
+  src/cuda_pointcloud_preprocessor/cuda_pointcloud_preprocessor.cu
+)
+
+target_link_libraries(cuda_pointcloud_preprocessor_lib
+  ${PCL_LIBRARIES}
+)
+
+target_include_directories(cuda_pointcloud_preprocessor_lib SYSTEM PRIVATE
+  ${autoware_point_types_INCLUDE_DIRS}
+  ${cuda_blackboard_INCLUDE_DIRS}
+  ${geometry_msgs_INCLUDE_DIRS}
+  ${nebula_common_INCLUDE_DIRS}
+  ${rclcpp_INCLUDE_DIRS}
+  ${rcl_interfaces_INCLUDE_DIRS}
+  ${sensor_msgs_INCLUDE_DIRS}
+  ${tf2_INCLUDE_DIRS}
+)
+
+# Targets
+ament_auto_add_library(cuda_organized_pointcloud_adapter SHARED
+  src/cuda_organized_pointcloud_adapter/cuda_organized_pointcloud_adapter_node.cpp
+)
+
+target_link_libraries(cuda_organized_pointcloud_adapter
+  ${CUDA_LIBRARIES}
+)
+
+ament_auto_add_library(cuda_pointcloud_preprocessor SHARED
+  src/cuda_pointcloud_preprocessor/cuda_pointcloud_preprocessor_node.cpp
+)
+
+target_link_libraries(cuda_pointcloud_preprocessor
+  ${CUDA_LIBRARIES}
+  cuda_pointcloud_preprocessor_lib
+)
+
+rclcpp_components_register_node(cuda_organized_pointcloud_adapter
+  PLUGIN "autoware::cuda_organized_pointcloud_adapter::CudaOrganizedPointcloudAdapterNode"
+  EXECUTABLE cuda_organized_pointcloud_adapter_node
+)
+
+rclcpp_components_register_node(cuda_pointcloud_preprocessor
+  PLUGIN "autoware::cuda_pointcloud_preprocessor::CudaPointcloudPreprocessorNode"
+  EXECUTABLE cuda_pointcloud_preprocessor_node
+)
+
+install(DIRECTORY launch config
+  DESTINATION share/${PROJECT_NAME}
+)
+
+ament_auto_package()
+
+# Set ROS_DISTRO macros
+set(ROS_DISTRO $ENV{ROS_DISTRO})
+if(${ROS_DISTRO} STREQUAL "rolling")
+    add_compile_definitions(ROS_DISTRO_ROLLING)
+elseif(${ROS_DISTRO} STREQUAL "foxy")
+    add_compile_definitions(ROS_DISTRO_FOXY)
+elseif(${ROS_DISTRO} STREQUAL "galactic")
+    add_compile_definitions(ROS_DISTRO_GALACTIC)
+elseif(${ROS_DISTRO} STREQUAL "humble")
+    add_compile_definitions(ROS_DISTRO_HUMBLE)
+endif()

sensing/autoware_cuda_pointcloud_preprocessor/README.md

@@ -0,0 +1,20 @@
+# autoware_cuda_pointcloud_preprocessor
+
+## Purpose
+
+The pointcloud preprocessing implemented in `autoware_pointcloud_preprocessor` has been thoroughly tested in autoware. However, the latency it introduces does not scale well with modern LiDAR devices due to the high number of points they introduce.
+
+To alleviate this issue, this package reimplements most of the pipeline presented in `autoware_pointcloud_preprocessor` leveraging the use of GPGPUs. In particular, this package makes use of CUDA to provide accelerated versions of the already establishes implementations, while also maintaining compatibility with normal ROS nodes/topics. <!-- cSpell: ignore GPGPUs >
+
+## Inner-workings / Algorithms
+
+A detailed description of each filter's algorithm is available in the following links.
+
+| Filter Name                       | Description                                                                                                                                  | Detail                                            |
+| --------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------- |
+| cuda_organized_pointcloud_adapter | Organizes a pointcloud per ring/channel, so that the memory layout allows parallel processing in cuda                                        | [link](docs/cuda-organized-pointcloud-adapter.md) |
+| cuda_pointcloud_preprocessor      | Implements the cropping, distortion correction, and outlier filtering (ring-based) of the `autoware_pointcloud_preprocessor`'s cpu versions. | [link](docs/cuda-pointcloud-preprocessor.md)      |
+
+## (Optional) Future extensions / Unimplemented parts
+
+The subsample filters implemented in `autoware_pointcloud_preprocessor` will have similar counterparts in this package.

sensing/autoware_cuda_pointcloud_preprocessor/config/cuda_pointcloud_preprocessor.param.yaml

@@ -0,0 +1,22 @@
+/**:
+  ros__parameters:
+
+    base_frame: base_link
+    use_imu: true
+    use_3d_distortion_correction: false
+    distance_ratio: 1.03
+    object_length_threshold: 0.1
+    num_points_threshold: 4
+
+    self_crop.min_x: 1.0
+    self_crop.min_y: 1.0
+    self_crop.min_z: 1.0
+    self_crop.max_x: -1.0
+    self_crop.max_y: -1.0
+    self_crop.max_z: -1.0
+    mirror_crop.min_x: 1.0
+    mirror_crop.min_y: 1.0
+    mirror_crop.min_z: 1.0
+    mirror_crop.max_x: -1.0
+    mirror_crop.max_y: -1.0
+    mirror_crop.max_z: -1.0

sensing/autoware_cuda_pointcloud_preprocessor/docs/cuda-organized-pointcloud-adapter.md

@@ -0,0 +1,41 @@
+# cuda_organized_pointcloud_adapter
+
+## Purpose
+
+The node `cuda_pointcloud_preprocessor` expects a 2D pointcloud where every row represents a single channel/ring and each row's points are in non decreasing azimuth order.
+
+To utilize the previously mentioned node, this node provides an adapter to convert standard flat pointclouds (`height` equals to 1) to the required 2D tensor.
+
+In addition, this node uses the `cuda_blackboard`, a cuda transport layer that enables a zero-copy mechanism between GPU and GPU memory for both input and output.
+
+## Inner-workings / Algorithms
+
+To create the required 2D tensor, this node iterates the input pointcloud sequentially, filling the output 2D tensor depending on the input point's channel.
+
+The output tensor's size is also estimated in this node, based on the largest `channel` value and the maximum number of points per channel observed so far.
+
+## Inputs / Outputs
+
+### Input
+
+| Name                 | Type                            | Description               |
+| -------------------- | ------------------------------- | ------------------------- |
+| `~/input/pointcloud` | `sensor_msgs::msg::PointCloud2` | Input pointcloud's topic. |
+
+### Output
+
+| Name                       | Type                                             | Description                              |
+| -------------------------- | ------------------------------------------------ | ---------------------------------------- |
+| `~/output/pointcloud`      | `sensor_msgs::msg::PointCloud2`                  | Processed pointcloud's topic             |
+| `~/output/pointcloud/cuda` | `negotiated_interfaces/msg/NegotiatedTopicsInfo` | Processed pointcloud's negotiation topic |
+
+## Parameters
+
+### Core Parameters
+
+{{ json_to_markdown("sensing/autoware_cuda_pointcloud_preprocessor/schema/cuda_pointcloud_preprocessor.schema.schema.json") }}
+
+## Assumptions / Known limits
+
+- This algorithm assumes that the input points will be in non-decreasing azimuth order (per ring).
+- This node expects that the input pointcloud is flat (`height` equals to 1) and follows the `autoware::point_types::PointXYZIRC` layout defined in the `autoware_point_types` package.

sensing/autoware_cuda_pointcloud_preprocessor/docs/cuda-pointcloud-preprocessor.md

@@ -0,0 +1,46 @@
+# cuda_pointcloud_preprocessor
+
+## Purpose
+
+This node implements all standard pointcloud preprocessing algorithms applied to a single LiDAR's pointcloud in CUDA.
+In particular, this node implements:
+
+- crop boxing (ego-vehicle and ego-vehicle's mirrors)
+- distortion correction
+- ring-based outlier filtering
+
+## Inner-workings / Algorithms
+
+As this node reimplements the functionalities of the CPU-version algorithms, please have a look at the documentations of [crop-box](../../autoware_pointcloud_preprocessor/docs/crop-box-filter.md), [distortion correction](../../autoware_pointcloud_preprocessor/docs/distortion-corrector.md), and [ring-based outlier filter](../../autoware_pointcloud_preprocessor/docs/ring-outlier-filter.md) for more information about these algorithms.
+
+In addition to the individual algorithms previously mentioned, this node uses the `cuda_blackboard`, a cuda transport layer that enables a zero-copy mechanism between GPU and GPU memory for both input and output.
+
+## Inputs / Outputs
+
+### Input
+
+| Name                      | Type                                             | Description                               |
+| ------------------------- | ------------------------------------------------ | ----------------------------------------- |
+| `~/input/pointcloud`      | `sensor_msgs::msg::PointCloud2`                  | Input pointcloud's topic.                 |
+| `~/input/pointcloud/cuda` | `negotiated_interfaces/msg/NegotiatedTopicsInfo` | Input pointcloud's type negotiation topic |
+| `~/input/twist`           | `geometry_msgs::msg::TwistWithCovarianceStamped` | Topic of the twist information.           |
+| `~/input/imu`             | `sensor_msgs::msg::Imu`                          | Topic of the IMU data.                    |
+
+### Output
+
+| Name                       | Type                                             | Description                              |
+| -------------------------- | ------------------------------------------------ | ---------------------------------------- |
+| `~/output/pointcloud`      | `sensor_msgs::msg::PointCloud2`                  | Processed pointcloud's topic             |
+| `~/output/pointcloud/cuda` | `negotiated_interfaces/msg/NegotiatedTopicsInfo` | Processed pointcloud's negotiation topic |
+
+## Parameters
+
+### Core Parameters
+
+{{ json_to_markdown("sensing/autoware_cuda_pointcloud_preprocessor/schema/cuda_pointcloud_preprocessor.schema.schema.json") }}
+
+## Assumptions / Known limits
+
+- The CUDA implementations, while following the original CPU ones, will not offer the same numerical results, and small approximations were needed to maximize the GPU use.
+- This node expects that the input pointcloud follows the `autoware::point_types::PointXYZIRCAEDT` layout and the output pointcloud will use the `autoware::point_types::PointXYZIRC` layout defined in the `autoware_point_types` package.
+- The input pointcloud must be a 2D tensor where each row represents a different channel/ring with its points in non-decreasing azimuth order. Invalid points should contain 0-values. The `cuda_organized_pointcloud_adapter` provides such a pointcloud.

sensing/autoware_cuda_pointcloud_preprocessor/include/autoware/cuda_organized_pointcloud_adapter/cuda_organized_pointcloud_adapter_node.hpp

@@ -0,0 +1,77 @@
+// Copyright 2024 TIER IV, Inc.
+//
+// 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 AUTOWARE__CUDA_ORGANIZED_POINTCLOUD_ADAPTER__CUDA_ORGANIZED_POINTCLOUD_ADAPTER_NODE_HPP_
+#define AUTOWARE__CUDA_ORGANIZED_POINTCLOUD_ADAPTER__CUDA_ORGANIZED_POINTCLOUD_ADAPTER_NODE_HPP_
+
+#include <autoware/point_types/types.hpp>
+#include <autoware/universe_utils/ros/debug_publisher.hpp>
+#include <autoware/universe_utils/system/stop_watch.hpp>
+#include <cuda_blackboard/cuda_adaptation.hpp>
+#include <cuda_blackboard/cuda_blackboard_publisher.hpp>
+#include <cuda_blackboard/cuda_pointcloud2.hpp>
+#include <rclcpp/rclcpp.hpp>
+
+#include <sensor_msgs/msg/point_cloud2.hpp>
+
+#include <tf2/transform_datatypes.h>
+#include <tf2_ros/buffer.h>
+#include <tf2_ros/transform_listener.h>
+
+#include <chrono>
+#include <deque>
+#include <memory>
+#include <vector>
+
+namespace autoware::cuda_organized_pointcloud_adapter
+{
+
+class CudaOrganizedPointcloudAdapterNode : public rclcpp::Node
+{
+public:
+  explicit CudaOrganizedPointcloudAdapterNode(const rclcpp::NodeOptions & node_options);
+  ~CudaOrganizedPointcloudAdapterNode() = default;
+
+private:
+  void estimatePointcloudRingInfo(
+    const sensor_msgs::msg::PointCloud2::ConstSharedPtr input_pointcloud_msg_ptr);
+
+  bool orderPointcloud(
+    const sensor_msgs::msg::PointCloud2::ConstSharedPtr input_pointcloud_msg_ptr);
+
+  // Callback
+  void pointcloudCallback(
+    const sensor_msgs::msg::PointCloud2::ConstSharedPtr input_pointcloud_msg_ptr);
+
+  // Subscriber
+  rclcpp::Subscription<sensor_msgs::msg::PointCloud2>::SharedPtr pointcloud_sub_{};
+
+  // CUDA pub
+  std::unique_ptr<cuda_blackboard::CudaBlackboardPublisher<cuda_blackboard::CudaPointCloud2>> pub_;
+
+  std::size_t num_rings_{0};
+  std::size_t max_points_per_ring_{0};
+
+  std::vector<std::size_t> next_ring_index_;
+  std::vector<autoware::point_types::PointXYZIRCAEDT> buffer_;
+  // autoware::point_types::PointXYZIRCAEDT * device_buffer_;
+  cuda_blackboard::CudaUniquePtr<std::uint8_t[]> device_buffer_;
+
+  std::unique_ptr<autoware::universe_utils::StopWatch<std::chrono::milliseconds>> stop_watch_ptr_;
+  std::unique_ptr<autoware::universe_utils::DebugPublisher> debug_publisher_;
+};
+
+}  // namespace autoware::cuda_organized_pointcloud_adapter
+
+#endif  // AUTOWARE__CUDA_ORGANIZED_POINTCLOUD_ADAPTER__CUDA_ORGANIZED_POINTCLOUD_ADAPTER_NODE_HPP_