Overview of Scenario Generation from Recorded Sensor Data

The Scenario Builder for Automated Driving Toolbox™ support package enables you to create virtual driving scenarios from vehicle data recorded using various sensors, such as a global positioning system (GPS), inertial measurement unit (IMU), camera, and lidar. To create virtual driving scenarios, you can use raw sensor data as well as recorded actor track lists or lane detections. Using these virtual driving scenarios, you can mimic real-world driving conditions and evaluate autonomous driving systems in a simulation environment.

Scenario generation from recorded sensor data involves these steps:

1. Preprocess input data.

2. Extract ego vehicle information.

3. Extract scene information.

4. Extract information of non-ego actors.

5. Create, simulate, and export scenario.

Preprocess Input Data

Scenario Builder for Automated Driving Toolbox supports a variety of sensor data. You can load recorded data from GPS, IMU, camera, or lidar sensors into MATLAB^®. You can also load processed lane detections and actor track list data to create a virtual scenario.

After loading the data into MATLAB, you can perform these preprocessing steps on sensor data:

• Align the recorded timestamp range of different sensors by cropping their data into a common timestamp range. The Scenario Builder for Automated Driving Toolbox support package supports timestamp values in the POSIX^® format.

• Normalize and convert the timestamp values into units of seconds.

• Organize the sensor data into formats that Scenario Builder for Automated Driving Toolbox supports. For more information, see Preprocess Lane Detections for Scenario Generation.

• Specify the region of interest (ROI) in the GPS data for which you want to create a scenario. Use the getMapROI function to get the coordinates of a geographic bounding box from the GPS data. To visualize geographic data, use the geoplayer object.

• Convert geographic coordinates to local Cartesian coordinates using the latlon2local function.

Extract Ego Vehicle Information

The local Cartesian coordinates that you obtain from the latlon2local function specify the ego waypoints. Because these waypoints are directly extracted from raw GPS data, they often suffer from GPS noise due to multipath propagation. You can smooth this data to remove noise and better localize the ego vehicle. For more information on smoothing GPS data, see Smooth GPS Waypoints for Ego Localization. Then, generate the ego trajectory from the waypoints and the corresponding time information using the waypointTrajectory (Sensor Fusion and Tracking Toolbox) System object™.

To improve road-level localization of the ego vehicle, you can fuse the information from GPS and IMU sensors. For more information, see Ego Vehicle Localization Using GPS and IMU Fusion for Scenario Generation. To get lane-level localization of the ego vehicle, you can use lane detections and HD map data. For more information, see Ego Localization Using Lane Detections and HD Map for Scenario Generation.

Extract Scene Information

To extract scene information, you must have road parameters and lane information. Use the roadprops function to extract road parameters from the desired geographic ROI. You can extract road parameters from these sources:

• ASAM OpenDRIVE^® file

• HERE HD Live Map¹ (HERE HDLM)

• OpenStreetMap^®

• Zenrin Japan Map API 3.0 (Itsumo NAVI API 3.0)²

The function extracts parameters for any road within the ROI. To generate a scenario, you need only the roads on which the ego vehicle is traveling. Use the selectActorRoads function to get the ego-specific roads.

Standard definition (SD) map data often lacks detailed lane information, which is essential for navigation in an autonomous system. You can use the updateLaneSpec function to get detailed lane information from recorded lane detections. To extract lane information from raw camera data, see Extract Lane Information from Recorded Camera Data for Scene Generation.

You can also create a scene from recorded lidar data. For more information, see Generate RoadRunner Scene from Recorded Lidar Data.

Extract Non-Ego Actor Information

After extracting ego information and road parameters, you must use non-ego actor information to create a driving scenario. Use the actorprops function to extract non-ego actor parameters from the actor track list data. The function extracts various non-ego parameters, including speed, yaw, and entry and exit times. For more information, see the actorprops function.

To extract an actor track list from camera data, see Extract Vehicle Track List from Recorded Camera Data for Scenario Generation. You can also extract a vehicle track list from recorded lidar data. For more information, see Extract Vehicle Track List from Recorded Lidar Data for Scenario Generation.

Create, Simulate, and Export Scenario

Create a driving scenario using a drivingScenario object. Use this object to add a road network and specify actors and their trajectories from your extracted parameters. For more information on how to create and simulate scenario, see Generate Scenario from Actor Track List and GPS Data.

You can export the generated scenario to the ASAM OpenSCENARIO^® file format using the export function of the drivingScenario object.

Using a roadrunnerHDMap object, you can also create a RoadRunner HD Map from road network data that you have updated using lane detections. The RoadRunner HD Map enables you to build a RoadRunner scene. For more information, see the Generate RoadRunner Scene from Recorded Lidar Data example.

You can export actor trajectories to CSV files, and generate RoadRunner scenario by importing CSV trajectories into RoadRunner Scenario. For more information, see Generate RoadRunner Scenario from Recorded Sensor Data.

You can create multiple variations of a generated scenario to perform additional testing of automated driving functionalities. For more information, see Overview of Scenario Variant Generation.