St Germain en Laye, October 1st 2024.
A NEW PARADIGM FOR AUTONOMOUS VEHICLE : reduce development, test/validation times by at least a factor 100.
| When we want to automate the driving of a vehicle, it is natural to contact automation engineers. Their approach is to list the use cases. Each use case is a scenario, for example: a red vehicle follow another one blue approaching a roundabout with also a bicycle traveling in front. The events building this scenario are: presence of a roundabout, presence of blue vehicle located in front, presence of a bicycle located in front. Depending on the relative and absolute speed of these vehicles and their position as well as that of the roundabout (N scenarios built from these events), we deduce the desired behavior for our red vehicle to be automated. Each desired behavior is obtained by a control law which generally includes 6 parameters. We therefore obtain 6 x N parameters just for this simple use case. This is to be multiplied by the number of use cases. In practice, we quickly end up with a few hundred or even a few thousand parameters, which is very difficult to adjust or even impossible. And the run time of such a system requires strong embedded CPUs. It is difficult to scale. |
 |
| Nexyad has developed a new approach based on a prudence metric that makes the solution scalable and which uses automation only to modulate the comfort parameters. This approach does not work on scenarios but on events: we assign to each event a « prudence function ». In our case mentioned above, we will characterize the following events; a roundabout, a vehicle in front, a bicycle in front. This makes a total of 3 « prudence functions » (instead of N scenarios). Once these 3 events have been characterized in terms of prudence, we take into account the current scenario by combining the « prudence functions » with fuzzy logic and theory of possibilities. All possible scenarios for combining these three events are taken into account in a logical manner. |
We keep only one parameter which is the « lack of prudence target », which characterizes the sportiness of the vehicle. Prudence function are independent functions that can be tunes separately. This new paradigm reduces times of development and test/validation by a factor 100 at least. It calculate prudence of automated vehicles (from predictive ACC to level 4). This prudence is a KPI of the vehicle. It is an eXplainable Artificial Intelligence (XAI). This system runs with only 5% of the CPU of a conventional PC type computer. The autonomous vehicle can at least be scalable and adjustable. This technology is called SafetyNex, to learn more: Autonomous Cars & Autonomous Trucks driven by Prudence (nexyad.net) |
#Nexyad #AI #ArtificialIntelligence #AutonomousVehicle #SelfDriving #PredictiveACC #FuzzyLogic #PossibilityTheory #Scalable #XAI #eXplainableArtificialIntelligence