Refine
Document Type
- Conference Paper (28)
- Article (2)
Institute
- Fakultät Elektro- und Informationstechnik (24)
- CARISSMA Institute of Automated Driving (C-IAD) (22)
- Fakultät Informatik (8)
- Human-Computer Interaction Group (HCIG) (8)
- CARISSMA (6)
- Fakultät Elektrotechnik und Informatik (2)
- AImotion Bavaria (1)
- CARISSMA Institute of Electric, Connected and Secure Mobility (C-ECOS) (1)
- Institut für Innovative Mobilität (IIMo) (1)
Review
- peer-review (22)
- nein (5)
- ja (2)
Version
- published (5)
Simulation-based test methods with an automotive camera-in-the-loop for automated driving algorithms
(2022)
Dynamic Vehicle-in-the-Loop
(2022)
In automated driving functions (ADF) testing, novel methods have been developed to allow the combination of hardware and simulation to ensure safety in usage even at an early stage of development. This article proposes an architecture to integrate an entire test vehicle—denominated Dynamic Vehicle-in-the-Loop (DynViL)—in a virtual environment. This approach enables the interaction of a real vehicle with virtual traffic participants. The vehicle is physically tested on an empty track, but connected to the CARLA simulator, in which virtual driving scenarios are created. The simulated environment is transmitted to the vehicle driving function which controls the real vehicle in reaction to the virtual objects perceived in simulation. Furthermore, the performance of the DynViL in different test scenarios is evaluated. The results show that the reproducibility of the tests with the DynViL is satisfactory. Furthermore, the results indicate that the deviation between simulation and DynViL variates according to the vehicle speed and the complexity of the scenario. Based on the performance of the DynViL in comparison to simulation, the DynViL can be implemented as a complementary test method to be added to the transition between hardware in the loop (HiL) and proving ground. In this test method, erratic or unexpected behavior generated by the driving function and controllers can be detected in the real vehicle dynamics in a risk-free manner.
Validity Analysis of Simulation-based Testing concerning Free-space Detection in Autonomous Driving
(2020)
Validation of a radar sensor model under non-ideal conditions for testing automated driving systems
(2022)
Automated driving functions (ADF) are considered as a potential solver of current problems in road traffic regarding safety, efficiency and comfort. However, testing ADF by naturalistic driving in the real world is subject to technical and ethical constraints. Virtual randomized controlled trial designs potentially contribute to bypass these limitations. For this purpose, real traffic is replaced by simulated traffic, constituting the “reference” in analogy to randomized controlled trials in medicine. Specific realizations of ADF can then be integrated into the simulated traffic as a “treatment” to evaluate their efficacy. A key challenge is modelling current manual traffic, taking into account stochastic variations in the cognitive and kinematic behavior of both drivers and vulnerable road users (VRU) such as pedestrians, cyclists, or e-scooter riders. Odd sample combinations of the underlying distributions can lead to accident risk and therefore have to be modeled realistically to generate validated efficacy estimates. In particular perceptual failures and degrading of perceived stimuli are regarded causal factors for failures in traffic, which is in general remarkably safe due to multiple redundancies. Therefore a model of human information acquisition constitutes an essential ingredient to our assessment paradigm. However, complex cognitive processes play a key role, which are themselves still under scientific investigation. What we do know is that inherent limited processing abilities of humans contribute to failures in the otherwise remarkably safe traffic flow process, especially in urban areas where cognitive demand is high. We therefore restrict ourselves to model the failure and degrading processes which ultimately lead to accident risk. The computational model we propose takes the limited processing capacity of humans into account and is suitable for the stochastic simulation of traffic scenarios.