How autonomous driving startup Wayve evaluates its own performance

November 25, 2020 | Case Study

As autonomous vehicles (AVs) progress to Level 4, a big question that remains is how to validate that an autonomous system is safe enough to deploy. It is extraordinarily difficult to evaluate a machine learning-based system in environments that are difficult to predict, let alone a safety-critical system. It's a bit easier to validate the individual components of an autonomy system. For example, AutoX's self-driving computer complies with ISO 26262's highest standard, deeming it fault-tolerant. Deepen creates a validation tool that assesses an AV's sensor fusion and perception capabilities to make sure they meet some minimum standard. However, as you zoom out to the full self-driving system, no standards or tools are universally used for validation.