SAFE-FM: Safety-Enhanced and Physics-Informed Foundation Models of Vehicle Behavior
| Reference number | |
| Coordinator | Chalmers Tekniska Högskola AB - Chalmers tekniska högskola Department of Computer Science and Engineering |
| Funding from Vinnova | SEK 5 038 538 |
| Project duration | July 2025 - June 2029 |
| Status | Ongoing |
| Venture | Safe automated driving – FFI |
| Call | Traffic-safe automation - FFI - spring 2025 |
Purpose and goal
The project aims to develop physics-informed and data-driven AI models that improve the safety, reliability, and efficiency of automated vehicles. By developing predictive, generative, and foundation models tailored for time-series data, the project will support advanced simulation, testing, and real-world deployment. This approach will address critical safety challenges, reduce costs, and enable scalable, explainable AI solutions aligned with industry needs and Vision Zero goals.
Expected effects and result
The project will develop predictive AI models for accurately forecasting vehicle behavior, generative tools for creating synthetic test cases covering critical safety scenarios, and a foundation model for analyzing and simulating time-series vehicle data. This will improve vehicle behavior modeling, reduce reliance on costly physical testing, increase the coverage of safety-critical scenarios, and integrate these models into real vehicle systems to prevent accidents and support Vision Zero.
Planned approach and implementation
The approach combines physics-informed machine learning, foundation models, and advanced generative techniques to model vehicle dynamics, predict vehicle dynamics, and simulate critical test scenarios. Models will be trained and validated using Volvo’s real-world data and test environments. A specialized foundation model for time-series data will be developed to unify analysis and generation tasks, with deployment planned in Volvo’s SIL/HIL frameworks for real-world validation.