feat(autoware_tensorrt_genad): Add TensorRT-accelerated GenAD end-to-end driving package

[YSH-research]

Checklist

• I've read the contribution guidelines.
• I've searched other issues and no duplicate issues were found.
• I've agreed with the maintainers that I can plan this task.

Description

Summary

Add a new package autoware_tensorrt_genad that integrates GenAD
(Generalized Autonomous Driving) as a TensorRT-accelerated
end-to-end planning module for Autoware.

Vision

The ultimate goal is to build an E2E + Rule-Based hybrid platform
in Autoware that is capable of obstacle avoidance, and to demonstrate
it in both CARLA simulation and real-vehicle driving.
This package is the first step toward that goal.

Background

• GenAD is a vision-based end-to-end autonomous driving model that performs perception and planning jointly.
• Training and validation have been completed on B2D dataset (GenAD_no_hlc variant).
• TensorRT quantization (FP16) is nearly complete.
• This follows the same pattern as the existing autoware_tensorrt_vad package.

Related

• Reference package: autoware_tensorrt_vad
• Model: GenAD paper

Purpose

Enable real-time end-to-end autonomous driving inference in Autoware by deploying GenAD with TensorRT acceleration. This provides an alternative E2E planning model alongside autoware_tensorrt_vad, with improved trajectory prediction performance validated on the B2D dataset.

Possible approaches

1. Follow the existing autoware_tensorrt_vad package structure as a reference implementation
2. Export GenAD (GenAD_no_hlc variant) to ONNX, then build TensorRT engine (FP16)
3. Wrap the inference pipeline in a ROS2 node with Autoware-compatible trajectory output interface
4. Reuse the DL4AGX pipeline for ONNX-to-TensorRT conversion

Definition of done

• autoware_tensorrt_genad package builds successfully in Autoware workspace
• TensorRT engine runs inference with FP16 precision
• ROS2 node publishes trajectory messages compatible with Autoware planning stack
• Launch files and README documentation are provided
• Basic performance benchmark (latency, FPS) is documented