Advanced driver assistance systems (ADAS) present a uniquely challenging development environment: an open-world infinite problem space, combined with a need to be repeatable to demonstrate safety and security. Current development approaches are inherently limited; handling rare events is critical, yet such events cannot be reproduced on demand in test systems. This presentation introduces the concept of deterministic ADAS development using a middleware that supports full validation and verification using recomputability. We will demonstrate how such middleware supports development through 'forensic recompute' and validation via 'credible recompute'. Additionally, we will illustrate how deterministic application performance can be enhanced using hardware accelerators.

• Why an open-world problem domain, such as ADAS, demands a non-traditional development approach
• How a reproducible system underpinned by deterministic middleware is the key to cracking ADAS verification and validation
• How 'forensic recompute' can assist in the debugging of a deterministic ADAS by guaranteeing reproducible behavior in the car and in the lab
• How 'credible recompute' enables validation of today’s ADAS and future systems using the same recorded data
• The performance implications of deterministic systems and some of the challenges when we try to interface with the non-deterministic world