Research questions UNECE regulation R79 specifies regulatory requirements for the purpose of type approval of Automatically Commanded Steering Functions (ACSF) including kinematic demands on the ego vehicle and the required safety distance to surrounding vehicles. UNECE regulation R157 defines how Lane Change Maneuvers (LCM) are incorporated into an Automatic Lane Keeping System (ALKS) as a part of automated driving. Safety distance requirements are defined, including the reasonable deceleration of and headway for an approaching rear target vehicle. The concrete rearward perception range depends on assumed maximum speeds and distances in the actual driving situation. The objective of this paper is to analyze real-world accidents and naturalistic driving data to derive typical driving parameters in critical and non-critical LCM. Relevant parameters including driving speeds, lane change timings and the maintained headways are analyzed in the first step. The second part of the paper determines the required perception spaces of an ego vehicle to conduct a safe LCM on motorways. The results are used to evaluate if a particular sensor field-of-view is sufficiently safe. Therefore, the criteria are based on commonly-experienced values rather than on potential worst-case assumptions. Methodology Accidents on motorways were investigated using GIDAS to obtain an overview and specifics of critical situations that led to crashes. Analyzing highD drone data, gathered by monitoring German motorways, the human driving behavior in lane following and lane-changing situations were investigated. The ego vehicle’s kinematics and timing before, during and after a typical LCM and the behavior of its surrounding vehicles were analyzed. Using an optimization algorithm, real-world data were applied to evaluate the minimum sensor field-of-view to achieve a variable degree of safety. This degree of safety would depend on the anticipated deceleration behavior of surrounding vehicles. Results Lane changes cause 8% of accident scenarios that a car might encounter on the motorway. 99% of the analyzed LCM in the highD drone data are completed within 5 seconds, according to the requirements in the R157. In 14% of LCM, the human driver kept a smaller safety distance to approaching vehicles in the target lane than was specified as the required critical distance in regulation R79. In GIDAS, the minimum rearward sensor perception range is 210 meters. In highD this distance is 110 meters. This assumes that no deceleration of the rear vehicle greater than 3 m/s² is acceptable during LCM. Limitations As a limitation to the study, it must be noted that, while GIDAS data is considered representative across Germany, the highD naturalistic driving data was collected using only a sample of motorway driving in Germany and may not be representative of all regions across Germany. Also, sensor range requirements are analyzed assuming ideal conditions and not considering perception errors and false activations. What is new The paper shows that theoretical worst-case assumptions for ego and rear vehicle speeds during LCM lead to unfeasible system layouts. Besides looking at edge cases, the focus of development should be on the most relevant speeds as observed in real-world data, both from traffic accidents and by measuring naturalistic driving. This underscores that rearward perception requirements for a safe, automated lane change system should be data-driven and based on real-world conditions.
Mr. Harald Feifel, Senior Expert, Continental Automomous Mobility
Deriving System Requirements for Automated Lane Changes from GIDAS and highD Data
FWC2023-SCA-017 • FISITA World Congress 2023 • Integrated safety, connected & automated driving
Upgrade your ICC subscription to access all Library items.
Congratulations! Your ICC subscription gives you complete access to the FISITA Library.
Retrieving info...
Available for purchase on the FISITA Store
OR