tensorrt_bevformer

Purpose

The core algorithm, named BEVFormer, unifies multi-view images into the BEV perspective for 3D object detection tasks with temporal fusion.

Inner-workings / Algorithms

Cite

• Zhicheng Wang, et al., “BEVFormer: Incorporating Transformers for Multi-Camera 3D Detection” [ref]
• This node is ported and adapted for Autoware from Multicoreware’s BEVFormer ROS2 C++ repository.

Inputs / Outputs

Inputs

Outputs

How to Use Tensorrt BEVFormer Node

Prerequisites

• TensorRT 10.8.0.43
• CUDA 12.4

Trained Model

Download the bevformer_small.onnx model to:

$HOME/autoware_data/ml_models/tensorrt_bevformer

Note: The BEVFormer model was trained on the nuScenes dataset for 24 epochs with temporal fusion enabled.

Test TensorRT BEVFormer Node with nuScenes

1. Integrate this package into your autoware_universe/perception directory.

2. To play ROS 2 bag of nuScenes data:

   cd autoware/src
   git clone -b feature/bevformer-integration https://github.com/naveen-mcw/ros2_dataset_bridge.git
   cd ..
   

Note: The feature/bevformer-integration branch provides required data for the BEVFormer.

Download nuScenes dataset and canbus data here.

Open and edit the launch file to set dataset paths/configs:

   nano src/ros2_dataset_bridge/launch/nuscenes_launch.xml
   

Update as needed:

   <arg name="NUSCENES_DIR" default="<nuScenes_dataset_path>"/>
   <arg name="NUSCENES_CAN_BUS_DIR" default="<can_bus_path>"/>
   <arg name="NUSCENES_VER" default="v1.0-trainval"/>
   <arg name="UPDATE_FREQUENCY" default="10.0"/>
   

1. Build the autoware_tensorrt_bevformer and ros2_dataset_bridge packages

   # Build ros2_dataset_bridge

   colcon build --packages-up-to ros2_dataset_bridge

   # Build autoware_tensorrt_bevformer

   colcon build --packages-up-to autoware_tensorrt_bevformer

   

File truncated at 100 lines see the full file

Changelog for package autoware_tensorrt_bevformer

0.53.0 (2026-09-29)

• Merge remote-tracking branch 'origin/main' into prepare-0.53.0-changelog

• refactor(perception): move node design files into each package (#13104) Co-authored-by: Claude Opus 5 (1M context) <<noreply@anthropic.com>>

• chore(pre-commit): update clang-format to v22.1.5 (#13126)

  • chore(pre-commit): update clang-format to v22.1.5

  * style(pre-commit): autofix ---------

• Contributors: Mete Fatih Cırıt, Ryohsuke Mitsudome, Taekjin LEE

0.52.0 (2026-06-30)

0.51.0 (2026-05-01)

• Merge remote-tracking branch 'origin/main' into tmp/bot/bump_version_base

• feat: default artifact paths to ~/autoware_data/ml_models (#12523) feat(launches,configs): default artifact paths to ~/autoware_data/ml_models Roll every per-package [data_path]{.title-ref} / [model_path]{.title-ref} launch-arg default from [$(env HOME)/autoware_data[/...]]{.title-ref} to [$(env HOME)/autoware_data/ml_models[/...]]{.title-ref} so standalone universe launches resolve artifacts under the new [~/autoware_data/ml_models/]{.title-ref} layout (autowarefoundation/autoware#7068). When invoked through autoware_launch the parent overrides cascade and already pin the new root (autowarefoundation/autoware_launch#1835); this commit closes the gap for users who launch a perception / localization / sensing / planning component directly with [ros2 launch <pkg>]{.title-ref}. 22 launch files updated (one-line default change each):

  • e2e/autoware_tensorrt_vad/launch/vad_carla_tiny.launch.xml
  • localization/yabloc/yabloc_pose_initializer/launch/yabloc_pose_initializer.launch.xml
  • perception/autoware_bevfusion/launch/bevfusion.launch.xml
  • perception/autoware_camera_streampetr/launch/streampetr.launch.xml
  • perception/autoware_image_projection_based_fusion/launch/pointpainting_fusion.launch.xml
  • perception/autoware_lidar_apollo_instance_segmentation/launch/lidar_apollo_instance_segmentation.launch.xml
  • perception/autoware_lidar_centerpoint/launch/lidar_centerpoint.launch.xml
  • perception/autoware_lidar_frnet/launch/lidar_frnet.launch.xml
  • perception/autoware_lidar_transfusion/launch/lidar_transfusion.launch.xml
  • perception/autoware_ptv3/launch/ptv3.launch.xml
  • perception/autoware_shape_estimation/launch/shape_estimation.launch.xml
  • perception/autoware_simpl_prediction/launch/simpl.launch.xml
  • perception/autoware_tensorrt_bevdet/launch/tensorrt_bevdet.launch.xml
  • perception/autoware_tensorrt_bevformer/launch/bevformer.launch.xml
  • perception/autoware_tensorrt_yolox/launch/{yolox_traffic_light_detector,yolox_tiny,yolox_s_plus_opt}.launch.xml
  • perception/autoware_traffic_light_classifier/launch/{car,pedestrian}_traffic_light_classifier.launch.xml
  • perception/autoware_traffic_light_fine_detector/launch/traffic_light_fine_detector.launch.xml
  • planning/autoware_diffusion_planner/launch/diffusion_planner.launch.xml

  - sensing/autoware_calibration_status_classifier/launch/calibration_status_classifier.launch.xml Drive-by README and test fixes:

  - e2e/autoware_tensorrt_vad/{README.md,docs/design.md}: also migrate the [$HOME/autoware_map/Town01]{.title-ref} examples to [$HOME/autoware_data/maps/Town01]{.title-ref}.

  - localization/yabloc/{README.md,yabloc_pose_initializer/README.md}: also migrate [$HOME/autoware_map/sample-map-rosbag]{.title-ref} to [$HOME/autoware_data/maps/demos/sample-map-rosbag]{.title-ref}.

  - control/autoware_smart_mpc_trajectory_follower/README.md: migrate the [map_path:=$HOME/autoware_map/sample-map-planning]{.title-ref} example to [$HOME/autoware_data/maps/demos/sample-map-planning]{.title-ref}.

  - simulator/autoware_carla_interface/README.md: migrate every [$HOME/autoware_map/Town01/...]{.title-ref} reference to [$HOME/autoware_data/maps/Town01/...]{.title-ref}.

  • perception/{autoware_bevfusion,autoware_image_projection_based_fusion,autoware_lidar_centerpoint,autoware_tensorrt_bevformer}/README.md: copy-paste examples updated to [~/autoware_data/ml_models/<pkg>]{.title-ref}.
  • perception/autoware_camera_streampetr/config/ml_package_camera_streampetr.param.yaml: header comment updated.
  • planning/autoware_diffusion_planner/README.md: prerequisites snippet updated.

  - sensing/autoware_calibration_status_classifier/test/{test_model_inference,test_calibration_status_classifier}.cpp: hardcoded fallback ONNX path updated. Users on the legacy layout can pin the old root with [data_path:=$HOME/autoware_data]{.title-ref} (or the per-package equivalent) on the command line. Refs:

File truncated at 100 lines see the full file