In the context of Hardware-in-the-Loop (HiL) test benches, the optical and acoustic evaluation of vehicle functions currently represents a major bottleneck in test automation. Conventional computer vision approaches are often rigid and highly sensitive to industrial disturbances such as specular reflections or changing lighting conditions. The SOVA project addresses this challenge by developing a modular, AI-based framework for physical end-to-end validation. It replaces manual evaluations and error-prone rule-based systems, enabling reliable assessment of the actual execution of vehicle functions.
The methodological foundation is a hybrid system architecture that integrates seamlessly into existing test infrastructures. To ensure economically efficient automation, SOVA combines three technological pillars:
Real-time state estimation: Deployment of robust AI methods for fast and reliable detection of optical signals and light sequences.
Kinematic analysis: Application of intelligent computer vision techniques for precise validation of mechanical and geometric processes.
Multimodal pattern recognition: Integration of image and audio analysis using modern AI models. This enables automated verification of complex vehicle feedback based on simple textual specifications—while significantly reducing the effort required to onboard new test scenarios.
This architecture bridges the gap between abstract test specifications and the complexity of real vehicle physics, enabling scalable, strategic, and cost-efficient automation.
The overarching objective of the project is to significantly reduce manual effort in test evaluation through the use of modular AI systems. Specifically, the following goals are pursued:
Overcoming optical disturbances: Reduction of misinterpretations in test evaluation by leveraging the superior robustness of AI against environmental influences in test bench conditions.
Efficiency enhancement: Automated real-time translation of physical measurements into unambiguous and audit-proof test verdicts.
Scalability & cost-efficiency: Significant reduction of configuration and training effort during vehicle changes through the use of flexible, generalizable AI approaches.
Standardization: Provision of a hardware-efficient, modular solution that can be consistently deployed across different test domains and test bench concepts.
