Fast generic sensor models for testing highly automated vehicles in simulation

Abstract

Automated driving is one of the big trends in automotive industry nowadays. Over the past decade car manufacturers have increased step by step the number and features of their driving assistance systems. Besides technical aspects of realizing a highly automated vehicle, the validation and approval poses a big challenge to automotive industry. Reasons for this are grounded in the complexity of interaction with the environment, and the difficulties in proving that an automated vehicle shows at least the same performance as a human driver, with respect to safety. There is a common agreement within the community, that physical vehicle testing only, will practically not be feasible for validation. Simulation is seen as a key enabler for validation as well as for development of highly automated vehicles. Several commercial simulation tools are available to model and test the entire chain of "sens -plan -act". But still there is a lack of fast generic sensor models, for the use from development to vehicle in the loop tests. This article proposes a fast generic sensor model to be used in testing and validation of highly automated driving functions in simulation.

Performance evaluation of a state-of-the-art automotive radar and corresponding modeling approaches based on a large labeled dataset

“…The flow chart in Figure 1 illustrates the data flow of a virtual test environment for ADAS/AD functions, including the presented object-based radar model. An environment simulation, e.g., Vires VTD (VIRES, 2019), IPG CarMaker (IPG, 2019), CARLA (Dosovitskiy et al, 2017), or AirSim (Shah et al, 2017), provides the test scenario and forwards the true state of the environment, called ground-truth, to the radar model. The radar model modifies the ground-truth according to the sensing capabilities of the considered radar.…”

“…Examples for generic sensor models that can simulate automotive radar are Hanke et al (2015); ; Muckenhuber et al (2019); Stolz and Nestlinger (2018). Hanke et al (2015) suggest to modify the ground-truth object list sequentially in a number of modules, each representing a specific sensor characteristic or environmental condition.…”

Development and Experimental Validation of an Intelligent Camera Model for Automated Driving

“…Low-fidelity sensor models that can simulate automotive cameras were given by Hanke et al [24], Muckenhuber et al [25], Schmidt et al [26], Stolz and Nestlinger [27]. Hanke et al [24], Schmidt et al [26] suggested modifying the ground-truth object list sequentially in a number of modules, and each module shall represent a specific sensor characteristic or environmental condition.…”

“…Hanke et al [24], Schmidt et al [26] suggested modifying the ground-truth object list sequentially in a number of modules, and each module shall represent a specific sensor characteristic or environmental condition. Stolz and Nestlinger [27] introduced a computationally efficient method to exclude all objects outside the sensor's FOV. Muckenhuber et al [25] presented a generic sensor model taking coverage, object-dependent fields of view, and false negative/false positive detections into account.…”

Virtual Risk Assessment for the Deployment of Autonomous Shuttles

“…The actual state of the virtual environment (ground truth) is forwarded to the sensor model, where it is modified based on the sensor model category (20). These are: Ideal sensor models: These models generate objectlist data based on perfect ground truth data from the virtual environment (21). Probabilistic sensor models: These models establish a probabilistic relationship between the sensor output and the ground truth of the virtual environment, for example, by adding statistical failure rates (22).…”