Bringing Real World to Simulation for Virtual Testing of Automated Driving (AD)

Automated driving spans a wide range of automation levels, from advanced driver assistance systems (ADAS) to fully autonomous driving. As the level of automation increases, so does the uncertainty for operational environments, increasing the needs for virtual validation. Automotive companies typically have an abundance of real-world data recorded from a vehicle that is suitable for open-loop simulations. However, recorded data is often not suitable for testing closed-loop control systems since it cannot react to changes in vehicle movement. Creating scenarios from recorded vehicle data is complicated and involves multiple steps, from sensor selection and mounting and calibration to data collection, visualization, labeling, and working with maps. In comparison with US or European traffic and road conditions, Indian traffic conditions have different scenarios on the road, e.g., different types of vehicle classes, pet and farm animals in addition to pedestrians, bullock carts, tractors, etc. This leads to imminent requirements for virtual validation with real-world scenarios and the need to recreate scenarios from the recorded vehicle data. See a methodology to recreate virtual driving scenarios from recorded vehicle data to enable closed-loop simulation for testing ADAS functionalities.  

A virtual driving scenario is created by recreating roads using GPS sensor data and map import from OpenStreetMap®, and targets vehicles using the detections from automatic labeling of lidar point cloud data. To test an ADAS feature in closed loop, one must model the ego vehicle (sensor and dynamics) as well as the scenario (roads and target vehicles). In the scenario, the driver’s vehicle is referred to as the ego vehicle and other vehicles on the road are referred to as target vehicles. The created virtual driving scenario is integrated into a closed-loop simulation to assess and test the behavior of ADAS features.