adas

Structured AI and Rulebooks

Structured AI and Rulebooks

A foundational aspect of driving – for both human-piloted and autonomous vehicles – is understanding the rules of the road. In ideal scenarios, vehicles and other road users obey all relevant rules without conflict. For example, a driver will commonly take the right of way at a green light, or avoid a lane change across solid lane markings.

In real-world driving conditions, however, conflicts frequently arise between rules, particularly in urban environments. A vehicle may be required to stop at a green light in order to avoid a jaywalking pedestrian, or a double-parked truck may necessitate a detour across solid lane markings. In such scenarios, “common practice” rules are followed, while other rules are deprioritized, to ensure safe and efficient traffic flow.

When rules conflict – that is, when safety, common practice, cultural norms, or ethical decisions take priority over standard road rules – how should autonomous vehicles behave? The answer: Just as a safe human driver would.

To achieve consistent driving behavior in such instances, Aptiv’s approach, called Structured AI, is a rigorous system using rich data collection and machine learning to encode logical descriptions of driving rules and preferences. A critical element of Structured AI is Aptiv’s Rulebooks, which prioritizes these rules, common practices, and all manners of driving preferences to ensure the safety and comfort take priority above all else.

Such a system allows an AV to determine the safest possible action, as it is it is impossible to satisfy all rules of the road due to the sometimes unpredictable actions of other road users. An AV Rulebook can vary from country to country, enabling developers to scale AVs globally without re-writing (or re-training) decision-making systems.

Currently no other development approach prioritizes rules as rigorously as Aptiv’s Structured AI – which means most AV operators are unable to bring service to multiple, varying markets because they simply cannot deploy safely at scale. By creating a Structured AI approach and developing an AV Rulebook, Aptiv is breaking down the barriers, giving more people around the world an opportunity to experience autonomous mobility.

In summary, foundational rules derived mathematically and defined by our team determine the safest possible action with minimal violation. This structure can serve as a guide for public officials and designated governing bodies in the creation of regulation that will ensure safe development and deployment. Standardization of safe AV driving principles is imperative, but that’s not to say every line of code must be uniform. AV’s “driving style” will vary based on manufacturer, similar to human driving variances. However, like human drivers, they must abide by an agreed set of prioritized guiding principles.

Recently, Emilio Frazzoli, Chief Scientist, Aptiv Autonomous Mobility team, published an academic paper on this topic, titled Liability, Ethics, and Culture-Aware Behavior Specification using Rulebooks. Dr. Frazzoli’s paper can be read here.

A foundational aspect of driving – for both human-piloted and autonomous vehicles – is understanding the rules of the road. In ideal scenarios, vehicles and other road users obey all relevant rules without conflict. For example, a driver will commonly take the right of way at a green light, or avoid a lane change across solid lane markings.

In real-world driving conditions, however, conflicts frequently arise between rules, particularly in urban environments. A vehicle may be required to stop at a green light in order to avoid a jaywalking pedestrian, or a double-parked truck may necessitate a detour across solid lane markings. In such scenarios, “common practice” rules are followed, while other rules are deprioritized, to ensure safe and efficient traffic flow.

When rules conflict – that is, when safety, common practice, cultural norms, or ethical decisions take priority over standard road rules – how should autonomous vehicles behave? The answer: Just as a safe human driver would.

To achieve consistent driving behavior in such instances, Aptiv’s approach, called Structured AI, is a rigorous system using rich data collection and machine learning to encode logical descriptions of driving rules and preferences. A critical element of Structured AI is Aptiv’s Rulebooks, which prioritizes these rules, common practices, and all manners of driving preferences to ensure the safety and comfort take priority above all else.

Such a system allows an AV to determine the safest possible action, as it is it is impossible to satisfy all rules of the road due to the sometimes unpredictable actions of other road users. An AV Rulebook can vary from country to country, enabling developers to scale AVs globally without re-writing (or re-training) decision-making systems.

Currently no other development approach prioritizes rules as rigorously as Aptiv’s Structured AI – which means most AV operators are unable to bring service to multiple, varying markets because they simply cannot deploy safely at scale. By creating a Structured AI approach and developing an AV Rulebook, Aptiv is breaking down the barriers, giving more people around the world an opportunity to experience autonomous mobility.

In summary, foundational rules derived mathematically and defined by our team determine the safest possible action with minimal violation. This structure can serve as a guide for public officials and designated governing bodies in the creation of regulation that will ensure safe development and deployment. Standardization of safe AV driving principles is imperative, but that’s not to say every line of code must be uniform. AV’s “driving style” will vary based on manufacturer, similar to human driving variances. However, like human drivers, they must abide by an agreed set of prioritized guiding principles.

Recently, Emilio Frazzoli, Chief Scientist, Aptiv Autonomous Mobility team, published an academic paper on this topic, titled Liability, Ethics, and Culture-Aware Behavior Specification using Rulebooks. Dr. Frazzoli’s paper can be read here.

How helpful was this article?
i

 

×

Please let us know how helpful this article was, so we can provide you with the best content possible. If you have more feedback to share, please feel free to contact us.
Thank you!

Careers


Shape the future of mobility. Join our team to help create vehicles that are safer, greener and more connected.

View Related Jobs

Subscribe


All Attachments (1)