Scalable and agnostic safety frameworks

DriveSafeAI will develop the evidence base for an AI safety assurance framework that is agnostic to the AI system and scalable across a variety of Operational Design Domains (ODDs). Pioneering the novel OASISS (ODD-based AI Safety In Self-Driving Systems) approach, these methods can be easily adapted to new ODDs, ensuring straightforward verification and validation of AI-based software across different regions and deployment geographies—critical for deploying assisted and automated driving safely around the world.

The methodologies developed will benefit the entire self-driving industry, supporting all system architectures that use AI, either in discrete components or fully end-to-end.

Novel simulation techniques for scenario testing

Using Wayve’s novel simulation techniques, DriveSafeAI will develop realistic and diverse edge-case scenarios that are challenging to test in real-world conditions. An ODD and behaviour-based scenario training and testing approach will be created, enabling the practical testing and development of DriveSafeAI’s methods using simulation methods. 

Simulation is crucial to test the safety and performance of Assisted and Automated Driving Systems. Simulation is key to unlocking the breadth and diversity of scenarios that are necessary for putting these systems to the test—scenarios that a vehicle rarely experiences in on-road testing. Proving the correlation between simulated testing results and those you would expect to achieve in the real world is a topic of ongoing research across the industry. DriveSafeAI seeks to address a conceptual gap—developing an evidence base to inform guidance and best practice.

Below are examples of simulated driving scenes created by Wayve’s PRISM-1, scene reconstruction model.