feat(project): add Sparse4D

[ktro2828]

Description

This pull request introduces a new ROS 2 package, sparse4d, which scaffolds the integration of the Sparse4D detector with ROS 2. The package includes the CMake build system, launch files, configuration, and initial implementation of the detector and node classes. Additionally, it adds utility improvements to the TensorRT common interface and refactors transform broadcasting in the nuscenes dataloader for efficiency and clarity.

Sparse4D ROS 2 Package Introduction and Core Implementation:

• Added the sparse4d ROS 2 package, including CMakeLists.txt, package.xml, and a README describing its purpose and usage. The package is scaffolded to provide a node, launch file, and parameter configuration for Sparse4D integration, with the detector currently returning empty outputs as a placeholder. [1] [2] [3]
• Implemented the core detector and node classes in include/sparse4d/sparse4d_detector.hpp and include/sparse4d/sparse4d_node.hpp, defining configuration structures, inference interfaces, and ROS 2 node logic for image/camera info subscription, synchronization, and output publishing. [1] [2]
• Added a launch file (sparse4d.launch.xml) for easy startup, and a YAML parameter file for configuring ONNX model path, detector parameters, and input topics. [1] [2]

TensorRT Utility Enhancements:

• Extended the TrtCommon class to provide methods for querying tensor data types and IO modes by both tensor index and name, improving the flexibility and debuggability of TensorRT engine interactions. [1] [2]

Nuscenes Dataloader Refactor:

• Refactored the nuscenes dataloader's transform broadcasting logic to batch static transforms and introduce a _create_transform utility, improving code clarity and efficiency in broadcasting /tf_static transforms. [1] [2] [3]

How was this PR tested?

Notes for reviewers

None.

Effects on system behavior

None.