Tuning planning#

In this section, we will update, evaluate, and fine-tune the Autoware planning modules, with a specific emphasis on the lane driving modules within the YTU campus environment. The Autoware planning side consists of the following main lane driving sections: Behavior planning and motion planning. Our focus will be on fine-tuning these modules to enhance our planning performance in the campus environment.

Planning parameter tuning#

Behavior planning tuning#

Behavior velocity planner#

The Behavior velocity planner is a planner that adjusts velocity based on traffic rules. It loads modules as plugins. Please refer to the package documentation for more information about these modules.

To enable or disable Behavior velocity planner modules, we will enable or disable the necessary plugin modules in the default_preset.yaml file according to our preferences.

• For example, in the YTU campus environment, there are many speed bumps, pedestrians, and specific areas where autonomous driving is not permitted. We will enable the speed bump module to handle these conditions. The out of lane module also helps in these areas. In Autoware 2.0, it is a module of the motion velocity planner, and launch_out_of_lane_module is already true by default.

To enable a module, we need to set the launch_ argument to true for this module in the default_preset.yaml parameter file:

  - arg:
      name: launch_speed_bump_module
-     default: "false"
+     default: "true"

Speed bump module tuning#

• Our vehicle's cruising speed is set to 15 km/h. Therefore, we will decrease the default speed bump velocity limits in the speed_bump.param.yaml to allow for minimum and maximum speeds when going over speed bumps.

- min_speed: 1.39  # [m/s] = [5 kph]
+ min_speed: 1.11  # [m/s] = [4 kph]
- max_speed: 2.78  # [m/s] = [10 kph]
+ max_speed: 2.22  # [m/s] = [8 kph]

• Also, we will increase the slow_start_margin parameter to provide a greater margin for slowing down when approaching a speed bump. Please refer to the speed bump module page for more information about these parameters. It is essential to fine-tune these modules based on your specific vehicle and environmental conditions.

- slow_start_margin: 1.0
+ slow_start_margin: 2.0

• The following image illustrates the virtual wall created by the slow start margin of the speed bump module. If you increase or decrease the slow_start_margin parameter, you will observe that the position of the virtual wall changes is relative to the speed bump.

Avoidance#

The Avoidance module plays a pivotal role in Autoware's behavior planning scene modules, offering rule-based avoidance capabilities. It provides the flexibility to define behavior based on intuitive parameters like lateral jerk and avoidance distance margin, enabling adaptable avoidance strategies tailored to your specific environment. This module operates within lane constraints, requiring access to lane structure information to ensure compliance with traffic rules. For example, it triggers indicator signals when the vehicle crosses a lane. The key distinction between motion and behavior modules in the planning stack is their consideration of traffic rules. Please refer to the Avoidance Module page for more information about the module's capabilities.

We will modify and update certain avoidance rules and margin parameters to handle with the specific conditions of our YTU campus environment and the capabilities of our vehicle.

• First, we will disable the use of opposite lanelets while avoiding, as we do not want to utilize them in our YTU Campus environment. The default avoidance behavior in our environment is as depicted in the image below.

• To disable the use of the opposite lanelet in the avoidance, we will set use_lane_type to same_direction_lane in the static_obstacle_avoidance.param.yaml file:

  avoidance:
-   use_lane_type: "opposite_direction_lane"
+   use_lane_type: "same_direction_lane"

Now, we expect the avoidance module not to utilize the opposite lane in our area. If the vehicle cannot fit past an obstacle in the same lanelet, it will come to a stop, as depicted in the image below:

Since we disabled the use of opposite lanelets for the Avoidance module, there may be instances where we cannot avoid objects due to the shifted path not fitting within the available lanelet. As a small vehicle, we will reduce the soft_margin parameter to decrease the distance between objects and the ego vehicle. You can adjust these changes to any target object according to your preference. For more information on these parameters, please refer to the avoidance module documentation.

  avoidance:
    target_object:
      car:
        lateral_margin:
-         soft_margin: 0.5                            # [m]
+         soft_margin: 0.3                            # [m]

• Also, you can choose which objects can be avoided along the path. In the YTU campus environment, as mentioned earlier, there are many pedestrians, and we do not want to perform avoidance maneuvers for them. Therefore, we will disable the avoidance maneuver for pedestrians by modifying the target value in the static_obstacle_avoidance.param.yaml file. You can disable any target object according to your preference.

  avoidance:
    target_filtering:
      target_type:
-       pedestrian: true
+       pedestrian: false

Motion Planning Tuning#

Path optimizer#

The path optimizer generates a kinematically feasible and collision-free trajectory based on the input path and drivable area. It updates the trajectory's position and orientation but retains the velocity from the input path. Please refer to the Path Optimizer and Model Predictive Trajectory (MPT) documentation page for more information about this package.

• The YTU Campus environment features numerous U-turns, narrow roads, and roundabouts. Given our vehicle's maximum steering angle is insufficient for these conditions, we must consider our steering angle limit (defined in the vehicle_info.param.yaml) to navigate these road types without exceeding road boundaries. Therefore, we will enable the steer_limit_constraint parameter for the path optimizer:

  mpt:
    option:
-     steer_limit_constraint: false
+     steer_limit_constraint: true

• Additionally, we will modify our mechanism for checking outside the drivable area. By default, the path optimizer checks the four corner points of the MPT footprint. However, this may lead to incorrect information in some situations, as shown in the following image.

• To address this issue, we will enable the use_footprint_polygon_for_outside_drivable_area_check parameter to consider the footprint as a polygon and check if it exceeds the Lanelet2 boundaries.

  option:
-   use_footprint_polygon_for_outside_drivable_area_check: false # If false, only the footprint's corner points are considered.
+   use_footprint_polygon_for_outside_drivable_area_check: true # If false, only the footprint's corner points are considered.

Obstacle stop module#

In Autoware 2.0, the motion velocity planner has three modules for obstacles: the obstacle stop module, the obstacle slow down module, and the obstacle cruise module. The obstacle stop module inserts stop points into the trajectory for the detected objects. It can also stop for the input point cloud, which adds safety, because the point cloud also includes obstacles that the perception does not detect. By default, the point cloud check of the obstacle stop module is off. The obstacle cruise module uses only the dynamic objects.

The default_preset.yaml file enables or disables each module:

  - arg:
      name: launch_obstacle_stop_module
      default: "true"
  - arg:
      name: launch_obstacle_slow_down_module
      default: "true"
  - arg:
      name: launch_obstacle_cruise_module
      default: "true"

For more information on the parameters and inner working algorithms of the obstacle stop module, please refer to the documentation page.

In our campus environment, we want the vehicle to stop also for the point cloud. So we turn on the point cloud check of the obstacle stop module:

      obstacle_filtering:
        check_inside:
          default: true
-         pointcloud: false
+         pointcloud: true

In our YTU environment and test vehicle conditions, due to the short vehicle width, the trajectory is also short. To improve the detection range of the obstacle stop module, we will increase the lateral margin for point cloud obstacles. This margin must not be larger than the lateral_margin of pointcloud_preprocessing in motion_velocity_planner.param.yaml (2.0 m by default), because the planner removes the points outside that margin first.

      obstacle_filtering:
        ...
        lateral_margin: # lateral margin between the obstacles and ego's footprint
          nominal:
            default: 0.3
            unknown: 0.3
-           pointcloud: 0.3
+           pointcloud: 0.6

The following image illustrates the motion behavior of the trajectory, considering the lateral margin and the stop margin parameters. It was taken with the obstacle_stop_planner package of an earlier Autoware version.

Autoware has a lot of changeable parameters and methods. The mentioned sample parameter tuning examples are just a small part of what is possible. For more information on the modules and their parameters, please refer to the Autoware Universe documentation.