CORVETTE 2
The CORVETTE 2 project advances the cognitive collection and processing of vehicle and environmental data to enable real-time detection and interpretation of traffic scenarios for individual vehicles and fleets. Building on the first CORVETTE project, it enhances onboard data capture and pre-processing, allowing anomalies and irregular situations to be identified during operation.
The project expands applications such as tunnel and weather detection and dashboard symbol recognition while integrating multimodal sensor data for richer context understanding. Using eye-tracking and data from experienced drivers, CORVETTE 2 supports automated labeling of traffic scenarios, enabling their interpretation with large language models and translation into simulation-ready formats.
CORVETTE 2 systematically develops robust, scalable machine learning methods for anomaly detection, multimodal data compression, and automated model generation for embedded systems. Through continuous validation in defined use cases, the project ensures practical applicability, preparing the developed methods for integration into AVL’s ecosystem.
Ultimately, CORVETTE 2 contributes to the development of intelligent, data-driven mobility services, supporting safer and more efficient vehicle operation through systematic, scalable use of live vehicle data.
Goals
CORVETTE 2 aims to enable real-time, intelligent interpretation of traffic scenarios by advancing onboard data processing and anomaly detection in vehicles and fleets. It expands detection applications while integrating multimodal sensor data, including video, audio, and eyetracking, to enhance context awareness. The project also focuses on automating the labeling and interpretation of driving data using advanced machine learning and large language models, ensuring outputs are usable for simulation and system validation. Additionally, CORVETTE 2 seeks to develop methods that are robust, scalable, and ready for practical deployment within AVL’s ecosystem, supporting the evolution of cognitive, data-driven mobility solutions.
Approach
To reach its goals, CORVETTE 2 uses an iterative development process, refining methods through test drives and fleet data. The project integrates additional sensors within a modular, container-based architecture, ensuring scalable deployment. By enhancing real-time, onboard data processing and leveraging eye-tracking combined with multimodal inputs, CORVETTE 2 enables semi-automated scenario labeling for machine learning model training. Large language models facilitate the interpretation of complex traffic situations, translated into simulationready formats for validation.
Expected and Achieved Results
CORVETTE 2 will deliver a robust, scalable onboard system capable of real-time, multimodal anomaly and novelty detection within vehicle sensor data. It will enable situationally adaptive machine learning models, supporting automated model generation and updates on embedded systems. The project demonstrates the practical application of advanced machine learning to real-world driving scenarios, validated through targeted use cases. Key outcomes include automated scenario labeling, multimodal data compression for efficient storage, and seamless integration within AVL’s ecosystem. Scientifically, CORVETTE 2 advances machine learning for industrial applications, addressing challenges of robustness, privacy, and domain-specific processing. Results will be shared through publications and conferences, contributing to the broader scientific community. Overall, the project establishes a strong foundation for cognitive, data-driven services in vehicle and fleet environments, paving the way for safer, smarter, and more efficient mobility.
