In Safe Hands: A New Gold Standard of Safety Validation
Yaak is building a global network of 10,000+ local experts to train a Generative Safety AI that can score the level of safety, detect driving mistakes, and explain why it was a mistake with unprecedented accuracy; all without being told what a stop sign, car, or roundabout is. It simply learns by observing.
Driver training education in the EU follows a human-centric, safety-first, and rigorous methodology. At the core of it is a community of 225K trained professionals in the EU, certified to train new drivers. Collectively, these driving instructors have decades of experience in detecting, scoring, and correcting human driving mistakes.
Germany, historically, has one of the toughest driver training and driving exam standards. In the bigger German cities, licensed instructors log over 40 hours of practical driving lessons with each student, teaching them the rules of the road, safe operation of the vehicle, and—more importantly—providing rich contextual feedback on their driving.
This 360° video shows a subtle mistake like wrong positioning. Watch the upper middle front cam to see how the student drives too far into the opposite lane.
The standard of German driving lessons is matched only by the formal training which the driving instructors have to undertake to be certified. Spread over a period of 18 months, the topics range from driving physics, attaining knowledge of regional, dangerous training routes, and understanding operations and shortcomings of driver assistance and ADAS functions.
Equipped with years of in-depth knowledge of their local region, dangerous and difficult driving routes, and common driving mistakes, driving instructors catch both subtle and severe driving mistakes with ease. These data points are paramount for them to communicate what the student needs to improve.
360º video. Front cam: A student overlooking a right-of-way sign, so the instructor slams the brakes. Rear cam: A near, rear-end collision.
Apart from detecting and correcting mistakes, instructors run guided lessons which range from driving basics (1), parking basics (2), lane navigation (3), traffic interaction, night drives, highway drives, train crossings, hill drives, unprotected left turns (6), multi-lane left turns, interactions with cyclists etc.
Historically, driving instructors use paper-based tracking of student progress. With the Yaak app, this is now almost entirely automated. The instructor can now focus on running the driving lessons without the cognitive load of memorizing mistakes, lesson topics, and progress of the student.
Old school paper vs. tracking in the Yaak app
We accomplish this by installing eight cameras, an AI computer, 5G modems and a vehicle integration gateway on driving instructor vehicles. We record video and vehicle telematics data, and match that to a map via a deep integration. With this data, we can generate 360° videos, and visualize the driving lesson routes (and mistakes) with lesson topics encountered along the way (e.g. roundabouts, lane changes).
Driving instructors use the Yaak app to annotate driving mistakes during lessons. Yaak will soon add audio from driving lessons, as well as recording the correction and coaching information during lessons.
Instructors and examiners are becoming part of the AI future
Yaak was founded with the aim of building a platform and tools for the driving instructor community, as well as learn from their decades of experience in scoring, detecting, and correcting human driving.
At Yaak, we believe driving instructors and driving test examiners are local experts who know exactly how safe driving looks like in their cities. We believe they can—and should—play an important role in the effort to test and validate the increasingly complicated driving decisions automated vehicles are performing.
With 50+ Yaak-enabled vehicles already driving around Germany, we can collect +100,000 hours of curated drive data annually, and have a backlog of hundreds of vehicles for driving schools waiting for their Yaak kit—both in Germany and new countries.
The driving instructors operate the vehicle in two modes:
Student mode: Driving instructors annotate driving mistakes during lessons to track student progress, as well as to understand what routes to drive in upcoming lessons, and to assess when a student is ready for a driving test.
Expert mode: The instructors drive Yaak-enabled vehicles to generate expert drive data that helps Yaak understand how an expert would drive in their city.
Driving mistakes from three sample cities in Germany. From June 2023, Yaak is operating in +30 German cities.
Yaak is working towards building a new gold standard of safety validation. The network of experts and SafetyAI will be accessible to anyone working on safety critical products through an API.
In the next blog post we will be introducing SafetyAI. If you want to read why SafetyAI is a major evolution within safety for autonomous and human driving, follow us on LinkedIn and stay tuned for future blog posts.
