Smart MPC Trajectory Follower

Smart MPC (Model Predictive Control) is a control algorithm that combines model predictive control and machine learning. While inheriting the advantages of model predictive control, it solves its disadvantage of modeling difficulty with a data-driven method using machine learning.

This technology makes it relatively easy to operate model predictive control, which is expensive to implement, as long as an environment for collecting data can be prepared.

</a> </p> ## Requirements It's recommended to install these in a virtual environment. ```bash pip3 install numba==0.58.1 GPy ``` ## Provided features This package provides smart MPC logic for path-following control as well as mechanisms for learning and evaluation. These features are described below. ### Trajectory following control based on iLQR/MPPI The control mode can be selected from "ilqr", "mppi", or "mppi_ilqr", and can be set as `mpc_parameter:system:mode` in [mpc_param.yaml](./autoware_smart_mpc_trajectory_follower/param/mpc_param.yaml). In "mppi_ilqr" mode, the initial value of iLQR is given by the MPPI solution. > [!NOTE] > With the default settings, the performance of "mppi" mode is limited due to an insufficient number of samples. This issue is being addressed with ongoing work to introduce GPU support. To perform a simulation, run the following command: ```bash ros2 launch autoware_launch planning_simulator.launch.xml map_path:=$HOME/autoware_data/maps/sample-map-planning vehicle_model:=sample_vehicle sensor_model:=sample_sensor_kit trajectory_follower_mode:=smart_mpc_trajectory_follower ``` > [!NOTE] > When running with the nominal model set in [nominal_param.yaml](./autoware_smart_mpc_trajectory_follower/param/nominal_param.yaml), set `trained_model_parameter:control_application:use_trained_model` to `false` in [trained_model_param.yaml](./autoware_smart_mpc_trajectory_follower/param/trained_model_param.yaml). To run using the trained model, set `trained_model_parameter:control_application:use_trained_model` to `true`, but the trained model must have been generated according to the following procedure. ### Training of model and reflection in control To obtain training data, start autoware, perform a drive, and record rosbag data with the following commands. ```bash ros2 bag record /localization/kinematic_state /localization/acceleration /vehicle/status/steering_status /control/command/control_cmd /control/trajectory_follower/control_cmd /control/trajectory_follower/lane_departure_checker_node/debug/deviation/lateral /control/trajectory_follower/lane_departure_checker_node/debug/deviation/yaw /system/operation_mode/state /vehicle/status/control_mode /sensing/imu/imu_data /debug_mpc_x_des /debug_mpc_y_des /debug_mpc_v_des /debug_mpc_yaw_des /debug_mpc_acc_des /debug_mpc_steer_des /debug_mpc_X_des_converted /debug_mpc_x_current /debug_mpc_error_prediction /debug_mpc_max_trajectory_err /debug_mpc_emergency_stop_mode /debug_mpc_goal_stop_mode /debug_mpc_total_ctrl_time /debug_mpc_calc_u_opt_time ``` Move [rosbag2.bash](./autoware_smart_mpc_trajectory_follower/training_and_data_check/rosbag2.bash) to the rosbag directory recorded above and execute the following command on the directory ```bash bash rosbag2.bash ``` This converts rosbag data into CSV format for training models. > [!NOTE] > Note that a large number of terminals are automatically opened at runtime, but they are automatically closed after rosbag data conversion is completed. > From the time you begin this process until all terminals are closed, autoware should not be running. Instead, the same result can be obtained by executing the following command in a python environment: ```python from autoware_smart_mpc_trajectory_follower.training_and_data_check import train_drive_NN_model model_trainer = train_drive_NN_model.train_drive_NN_model() model_trainer.transform_rosbag_to_csv(rosbag_dir) ``` Here, `rosbag_dir` represents the rosbag directory. At this time, all CSV files in `rosbag_dir` are automatically deleted first. We move on to an explanation of how the model is trained. If `trained_model_parameter:memory_for_training:use_memory_for_training` in [trained_model_param.yaml](./autoware_smart_mpc_trajectory_follower/param/trained_model_param.yaml) is set to `true`, training is performed on models that include LSTM, and if it is set to `false`, training is performed on models that do not include LSTM. When using LSTM, cell states and hidden states are updated based on historical time series data and reflected in the prediction. The paths of the rosbag directories used for training and validation, `dir_0`, `dir_1`, `dir_2`,..., `dir_val_0`, `dir_val_1`, `dir_val_2`,... and the directory `save_dir` where you save the models, the model can be saved in the python environment as follows: ```python from autoware_smart_mpc_trajectory_follower.training_and_data_check import train_drive_NN_model model_trainer = train_drive_NN_model.train_drive_NN_model() model_trainer.add_data_from_csv(dir_0, add_mode="as_train") model_trainer.add_data_from_csv(dir_1, add_mode="as_train") model_trainer.add_data_from_csv(dir_2, add_mode="as_train") ... model_trainer.add_data_from_csv(dir_val_0, add_mode="as_val") model_trainer.add_data_from_csv(dir_val_1, add_mode="as_val") model_trainer.add_data_from_csv(dir_val_2, add_mode="as_val") ... model_trainer.get_trained_model() model_trainer.save_models(save_dir) ``` File truncated at 100 lines [see the full file](https://github.com/autowarefoundation/autoware_universe/tree/main/control/autoware_smart_mpc_trajectory_follower/README.md)

Changelog for package autoware_smart_mpc_trajectory_follower

0.53.0 (2026-09-29)

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: https://github.com/autowarefoundation/autoware/issues/7068

• style: update pre-commit (black 26.1.0, pre-commit-hooks-ros 0.10.2) (#12195)

• Contributors: Mete Fatih Cırıt, Taeseung Sohn, github-actions

0.50.0 (2026-02-14)

• Merge remote-tracking branch 'origin/main' into humble
• fix(smart_mpc_trajectory_follower): remove cmake pip install command (#12037)
• chore: reflect the move of the description packages (#11969)
• fix(smart_mpc_trajectory_follower): python3-torch into exec_depend with condition (#11942)

File truncated at 100 lines see the full file