Virtual testing of parking assistance systems on a real proving ground
- 自動車
- 先進運転支援システム (ADAS)
- 自動運転
- Over-the-Air Testing
02:39min
概要
Watch the video to learn about our advanced vehicle-in-the-loop test system, which combines real vehicle sensors with virtual environments. Our over-the-air simulation generates virtual objects in real time directly at the ultrasonic sensor level - the test vehicle perceives these as real. This allows complex parking maneuvers and scenarios to be tested safely, in a controlled manner, and repeatedly, offering significant advantages over traditional test drives.
The solution is also suitable for demanding applications such as virtual homologation tests in accordance with UN regulations, such as DCAS (UN ECE R171). This reduces costs and increases safety and reliability.
ホスト
Caius Seiger
Product Manager Sensor Simulation, dSPACE
FAQ
Q1: What is a virtual sensor, and why is it useful in automotive applications?
A1: A virtual sensor is a software-based component that estimates physical sensor outputs using data from other available sensors. It reduces hardware complexity and costs while maintaining system functionality, which is particularly beneficial for safety-critical automotive systems.
Q2: How are neural networks developed and integrated into TargetLink for virtual sensors?
A2: Neural networks are developed externally using frameworks like Keras or PyTorch, trained with relevant data, and exported as ONNX files. These ONNX models are then integrated into TargetLink using the adaptable behavior feature, which links the model with MATLAB scripts that generate the necessary C code via external code generator for ECU deployment.
Q3: What role do MIL, SIL, and PIL simulations play in this integration process?
A3: MIL (Model-in-the-Loop), SIL (Software-in-the-Loop), and PIL (Processor-in-the-Loop) simulations verify the neural network’s behavior at different stages - from initial training, through generated code, to embedded hardware execution - ensuring consistency and accuracy across environments.
Q4: How does TargetLink ensure that the integration of the neural network model remains up to date?
A4: TargetLink includes an automated up-to-date check that runs before code generation. It verifies if the neural network model and the generated implementation are synchronized, triggering regeneration of code if needed, which maintains alignment and reduces manual errors.
Q5: Can this integration approach support multiple target hardware platforms?
A5: Yes, using TargetLink’s build extension feature, developers can specify different source files for different processors or hardware platforms within the TargetLink Data Dictionary, allowing easy switching between targets without modifying the model or reconfiguring the project.
